Control Theory and Technology

https://doi.org/10.1007/s11768-021-00051-1

RESEARCH ARTICLE

Support optimal scheduling with weighted random forest

for operation resources
Li Li1 · Qingyun Yu1 · Haoyi Shi1 · Yuguang Liu1

Received: 14 October 2020 / Revised: 18 November 2020 / Accepted: 31 December 2020

© South China University of Technology, Academy of Mathematics and Systems Science, CAS and Springer-Verlag GmbH Germany, part of Springer
Nature 2021

Abstract
Operation-related resources are lots of manpower and material with the characteristics of high cost and high income in
hospitals, and scheduling optimization is a very important research issue in medical service. In this paper, to cope with
the actualities of operation resources scheduling, such as poor planning, lack of standardized scheduling rules, chaotic use
of the operating rooms, and many human interference factors, we propose a systematic approach to optimize scheduling
problems based on multiple characteristics of operating resources. We first design a framework that includes the composite
dispatching rules (CDR), optimization ideology, and feedback mechanism, in which the CDR integrates flexible operating
time, hold-up time of medical facilities, available time of medical staff, and multiple constraints. The optimization ideology
is carried out through a learning model based on the weighted random forest (WRF) algorithm. The feedback mechanism
enables the approach to realize closed-loop optimizations adaptively. Finally, the superiority of the systematic scheduling
approach (SSA) is analyzed through numerical experiments on a simulation platform. Results of the simulation experiments
show that the proposed scheduling method can improve performances significantly, especially in the waiting time of patients.

Keywords Scheduling · Operation resources · Composite dispatching rules · Feedback mechanism

1 Introduction As an important safeguard department of the hospital, the

operating room is a unit with various medical resources.
In recent years, medical institutions around the world have It not only has high-tech precision medical equipment and
been under tremendous pressure to improve efficiency, expensive medical supplies, but also has all kinds of rigor-
strengthen management and enhance their competitive- ously trained medical professionals. Thus, it has high human
ness [1]. Under the circumstance that the supply capacity of cost as well as high equipment cost [4, 5]. In addition, opera-
medical services is unlikely to improve rapidly in the short tion revenue is the most important source of income for a
term, one of the effective ways to alleviate the contradiction hospital, accounting for 2/3 of the total hospital revenue, and
between supply and demand and maximize the utilization the operating room accounts for 40% of the entire hospital
of public services is to optimize hospital resource manage- resource costs, which is an important facility resource with
ment and improve the efficiency of medical services [2, 3]. the characteristics of high cost and high revenue in hospitals
[6, 7].
The comprehensive coordination of surgical patients’
* Li Li
[email protected] demands on the operating rooms, physicians, and surgical
resources is critical to hospital services. Surgical schedul-
Qingyun Yu
[email protected] ing involves operating rooms, surgeons, anesthetists, nurses,
paramedics, and surgical equipment, to ensure the maximum
Haoyi Shi
[email protected] utilization of each surgical resource [8, 9]. The problem of
surgical scheduling can be divided into advance scheduling
Yuguang Liu
[email protected] and allocation scheduling. However, since the integer plan-
ning model for surgical scheduling is a nonlinear NP-hard
1
College of Electronics and Information Engineering, Tongji problem, it cannot be solved even as an approximate solution
University, Shanghai 201804, China

1Vol.:(0123456789)
3
 L. Li et al.

[10]. Although scheduling theory and methods have been directions in the development of surgical scheduling [19].
widely applied and maturely developed, they are only occa- Pham et al. [20] viewed their optimization problem as a
sionally applied to the service industry, and the scheduling multi-modal block processing workshop scheduling problem
theory in the medical service industry is in urgent need of and established a mixed integer correspondingly to reduce
improvement, which can be efficiently solved by industrial the impact of performance reduction (maximum completion
engineering methods and operations research [11, 12]. The time) on operating room timeouts. Based on their concept
operating resources, as the key in the medical service indus- of overflow block time on the traditional one, Dexter et al.
try, will be greatly improved if these theories and methods [21] proposed gave a method for allocating it, i.e., allocating
are used [13]. In this context, based on the idea of machine regular block time to each physician and allocating over-
learning and closed-loop optimization ideology, we propose flow block time to the entire medical group. Lamiri et al.
a systematic approach to achieve an efficient scheduling pro- [22] demonstrated the random optimization problem with
gram of operation resources. 3-partition, where they allocated patients on a schedule to
The arrangement of this paper is as follows. A survey of reduce patient-related costs and overtime costs in the oper-
the relevant literature is given in Sect. 2. Section 3 describes ating room. Wachtel et al. [23] proposed that the allocation
the problem and Sect. 4 presents a model formulation of of additional block time should not be based on the sur-
operation resources’ allocation. Then, Sect. 5 proposes the geon’s operating time utilization, but should be under the
resolving thoughts, approach framework, and two main consideration of the marginal benefit of the operation. To
research contents, including composite dispatching rules maximize the utilization of operation resources and mini-
(CDR) and parameter optimization. Following that, Sect. 6 mize the possibility of overtime by introducing slack time,
conducts a series of numerical experiments to study the Hans et al. [24] presented a priority-based structural heu-
scheduling performance of the proposed dispatching rules ristics method. Domestically, surgical scheduling research
and the systematic scheduling approaches. Finally, Sect. 7 is still in its infancy. Shu et al. [25] used a mathematical
concludes this paper and gives considerations for future model to optimize the operating room scheduling scheme to
work. maximize the operating room utilization, taking into account
the two-dimensional resource constraints of the operating
room and the surgical assistant nurse, and established an
2 Literature review object planning model to obtain the mathematically optimal
results. Zhong et al. [26] adopted an optimization model
A hospital is a service system that is constrained by lim- based on operations research, combinatorial optimization,
ited manpower and material resources. Operation resources, and scheduling theory for operation resources scheduling
characterized by high cost, are considered to be the key management. By reviewing and evaluating the research lit-
facilities of hospitals because they have high demand for erature on operation resources scheduling, Bai et al. [27]
expensive resources, such as personnel and equipment [14, introduced the latest development of surgical scheduling
15]. For hospitals, the operation is the largest source of rev- theory in foreign hospitals, its practical application, and
enue, and it has an enormous impact on the overall perfor- its research prospect in China. It can be seen that there are
mances of hospitals. Therefore, researchers have made sub- abundant research results on operating room scheduling, but
stantial efforts to improve performance related to operation how to better apply production scheduling theory to operat-
resources [16]. ing room scheduling optimization has rarely been reported.
Over the past 60 years, the research on the optimization Coupled with the constant changes in the medical environ-
of hospital operation resources scheduling has achieved ment, new research questions are emerging, and this issue
fruitful results [17]. Early on, studies reviewed the research needs to be studied in depth.
literature on operation resource scheduling based on patient Surgical scheduling must be considered to be dynami-
demands, and distinguished advance scheduling and alloca- cally adjustable, because the operation is different from
tion scheduling. That is, advance scheduling is the process of the processing of workpieces production scheduling field.
determining a certain patient’s operation date, while alloca- And the operating time of an operation may be greatly
tion scheduling needs to determine the operation resources deviated due to various influencing factors such as indi-
in which the operation will be performed on a certain day vidual patient differences or changes in patients’ condi-
and the time to start the operation [18]. Recently, studies tions. In addition, emergency operation may also affect
have reviewed the latest research literature related to surgi- the original scheduling strategy. Moreover, such changes
cal planning and scheduling, categorized the research results are often unpredictable in advance, and the occurrence
from the perspective of problem description and technical of such uncertainties will lead to the entire schedul-
characteristics, and proposed that the study of uncertainty ing results being affected, requiring timely adjustment
in operation resources scheduling will be one of the main according to the actual situation and giving a scheduling

13
Support optimal scheduling with weighted random forest for operation resources

strategy according to the new actual situation [28–31]. 3 Problem description

These aforementioned scheduling methods did not con-
sider interjection of emergency operations, though selec- The object of this study is the patient who is brought to the
tive patients can be re-scheduled [32]. Therefore, a sys- hospital and requires operation. The types of operation is
tematic scheduling approach (SSA) will be conducted to divided into 3 categories according to the severity of the
improve corresponding performances such as the utiliza- injury (the patient’s injury severity is divided into severe,
tion of operation resources, waiting time of patients, and medium, and light): major operation, medium operation, and
resource equalization. minor operation; the hospital’s medical resources include
In the past few years, both industry and academia have personnel resources (anesthetists, doctors, nurses, etc.) and
paid more and more attention to the application of data equipment resources (anesthesia rooms, operating rooms,
mining methods in performance indicator forecasting, and surgical equipment, inpatient ward, etc.). Medical resources
expect to seek statistical rules from the operation data of required for operation can be divided into three types accord-
daily activities to accurately predict the key performance ing to different functions: the preparation phase (anesthesi-
indicators under the current scheduling scheme [33, 34]. ologists, anesthetic room, etc.), treatment during operation
If these performance indicators can be controlled pre- (doctors, nurses, surgical equipment, etc.), and postopera-
cisely, it will help hospitals to formulate scientific and tive nursing and observation (nurses, inpatient ward, etc.).
effective daily production plans and optimize scheduling Accordingly, as Fig. 1 shows, the surgical service process is
strategies [35]. In the operation resource scheduling prob- divided into three stages: pre-operative, intra-operative, and
lems, data-based optimization has not been fully studied. post-operative. Each stage of service has a rest and waiting
In this research, we apply data-based methods, such as time before and after the service, and the time when the
random forest algorithm and some regression method, to post-operative is finished is called the completion time of
explore the potential relationship between performances the operation. One surgical stage corresponds to one type of
and operation status, and employ suitable dispatching operation resource, and the two stages and types of operation
parameters for superior scheduling performances [36, 37]. resources are independent of each other.
There have been abundant studies on operation sched- The description above is for selective operations. In
uling, but most of them did not consider the insertion of contrast to existing studies, we consider the insertion of
emergency operations. In addition, a growing number of emergency operations which will be described in detail in
machine learning methods are developed to mine poten- Sect. 5.1.
tial relationships among data for assisting decision-mak- The following assumptions are made to facilitate the dis-
ing. However, few studies succeed in combining advanced cussion of the proposed scheduling method:
machine learning methods with operation scheduling and
multiple constraints. Based on the analysis of existing (1) Focus on selective operation with probability of poten-
literature, we take emergency operations into considera- tial interruptions of emergency operation.
tion to make the scheduling problem more realistic. In (2) For selective operation, the first one starts at 8:00 am.
addition, we adopt a machine learning method to excavate (3) Operation resources and patients must be ready on time.
potential scheduling knowledge through relevant data and (4) Surgeons are not permitted to enter other operation
strive to generate the optimal scheme suitable for the cur- rooms or other inpatient rooms before finishing the
rent situation. current operation.

Anesthesia Operating Observation

room_1 room_1 room_1

Anesthesia Operating Observation

room_2 room_2 room_2
Patient buffer

Anesthesia Operating Observation

room_n room_n room_n

Fig. 1  Three stages of one operation

13
 L. Li et al.

(5) Surgical disinfection and cleaning are calculated within • Ti3: duration time of postoperative observation in opera-
the duration of operation. tion i.
• Cw : working hours in the wth day.
• TSw : start time of work in the wth day, w ∈ ℕ+.
• TEw : end time of work in the wth day, w ∈ ℕ+.
4 Model formulation
• A: an arbitrarily large integer.
• STij : start time of stage j of operation i in the original
This section expounds on a model formulation that aims at
scheme.
maximizing the utilization of operating resources and mini-
• ST′ij : start time of stage j of operation i in the latest
mizing the waiting time of patients. In model formulation,
scheme which consider the interruption of emergency
we also consider the potential probability in hospitals that
operations.
scheduling for selective operations was occasionally inter-
• 𝛿1: penalty coefficient of postponing in selective opera-
rupted due to incoming patients or emergencies requiring
tion.
immediate treatments. Details of the model are as follows:
• 𝛿2: penalty coefficient of postponing in emergency opera-
tion.
4.1 Variable definition • ETij : end time of stage j in operation i.
• cm
xi2 : a binary variable. If resource m (belongs to type C)
• I: set of operations (patients), i = 1, 2, … , n. was applied in the second stage of operation i, xi2 cm
= 1;
• ij : surgical stage of operation i (pre-operative, intra-oper- otherwise, xi2 = 0.
cm

ative, and post-operative), j ∈ {1, 2, 3}.

• C: set of different types of operation resources (anesthe- 4.2 Decision variable
sia rooms, anesthetists; operating rooms, doctors, surgi- {
cal equipment; inpatient ward, nurses). Km =
1, if resource m is a key resource,
• Mc : set of operation resources belongs to type C. 0, otherwise,
• nij : set of operation resources required for stage j of oper-
{
1, if operation i is a selective operation,
ation i. 𝜆i =
0, otherwise.
• Jcm : set of each operation stage which resource m
(belongs to type C) participate in.
• nCij : number of resources belong to type C in stage j of 4.3 Objective function
operation i.
• Tijcm : duration time of resource m (belongs to type C) on The scheduling of operating resources can be incorporated
stage j of operation i. into multi-objective optimization problems, to maximize the
• Ti1: duration time of preparatory stage in operation i. utilization of operation resources such as operating rooms and
• Ti2 : duration time of operation stage in operation i. surgeons, decrease the waiting time of patients while taking
into account the probability of receiving incoming patients
(emergency operation), and enhance resource equalization.
�∑� ���
STi − ST�i , 0 − 𝛿2 ⋅ 𝜆i − 1 ⋅ STi − ST�i (1)
�� � � � � �
f1 = min 𝛿1 ⋅ 𝜆i ⋅ max ,
i

N � �
Ti2cm ⋅ xi2
cm
⎛ ∑ ⎞
Mc �
⎜�∑ ⋅ Km ⎟
i=1
f2 = max ⎜
N N
⎟, (2)
⎜ m=1 �∑ � ⎟
cm ∑ cm
⎜ max ETi2 ⋅ xi2 ⋅ Km , ET(i−1)2 ⋅ xi2 ⋅ Km ⎟
⎝ i=1 i=1 ⎠

1 ∑ ∑ ∑ cm cm
⎛ T x ⎞
⎜ �M � m∈M j∈J i∈I ij ij ⎟
f3 = min max ⎜ �
� c� c i
⎟. (3)
c∈C ⎜
�
1
� 2⎟
∑ ∑ ∑ cm cm ∑ ∑ ∑ cm cm
⎜ ��Mc �� Tij xij − T x
j∈Ji i∈I
�M � m∈M j∈J i∈I ij ij ⎟
⎝ � c� c i ⎠

13
Support optimal scheduling with weighted random forest for operation resources

If someone resource was selected and the stage k precedes ETij − ETij ≥ Tijcm �
� , j ∈ Ji , j ∈ Ji , i ∈ I, c ∈ C, m ∈ Mc .
the stage l, Zkl = 1; otherwise, Zkl = 0. If operation i is the (6)
first operation in someday, ri = 1; otherwise, ri = 0 . Equa-
Equations (5) and (6) mean that the three stages of an opera-
tions (1) to (3) are three objective functions. Equation (1)
tion must be performed consecutively and the sequence can-
is intended to minimum the integrated waiting time of all
not be changed. The next stage can only be operated after
patients with different penalty coefficients. Equation (2) is
the previous stage is completed, and there should be no time
intended to maximize the ratio of active working time to
interval between adjacent stages.
the total working time. Equation (3) is the dispersion coef-
ficient of homogeneous resources, which aims to balance ETi� 2 − ETi2 + H 1 − Zi,i� ≥ Ticm � �
( )
� , i ∈ I, i ∈ I, i ≠ i ,
the utilization. (7)

4.4 Constraints ETi2 − ETi� 2 + HZi,i� ≥ Ti2cm , i ∈ I, i� ∈ I, i ≠ i� . (8)

Equations (7) and (8) illustrate that two stages of the same
Through a comprehensive survey on the relevant literature,
operation cannot be performed at the same time.
we can summarize constraints of operation scheduling into
three categories: process constraints, time constraints, and xijcm ⋅ xijcm� = 0. (9)
personnel constraints [38–40]. Process constraints mainly
mean that the three stages of operations must be carried Equation (9) illustrates that one resource can perform only
out following the established sequence. Time constraints one operation at the same time.
are used to ensure each stage of operations will be com- { }
pleted within the specified time, and the starting time of TSw ≤ STi1 ≤ TEw − Ti1 − Ti3 − max Ti2cm ⋅ xi2
cm
,
+ (10)
each stage should not be later than a set value. Personnel c ∈ Cij , m ∈ Mc , w ∈ N , i = 1, 2, … , n.
constraints mean that all operations are limited by finite
personnel. All these constraints will increase the difficulty Equation (10) means that each operation must be performed
of the operation schedule. in one day, i.e., the three stages of one operation are not
There exist some rare and expensive armamentariums allowed to span 2 days.
that are required, and the finite expensive armamentarium ∑ cm
xij = ncij , c ∈ Cij .
probably cannot cope with those operations. Therefore, the m∈M
(11)
c

reasonable arrangement of armamentarium is necessary

even this will make the scheduling problem more difficult Equation (11) illustrates that all the resources involved in the
to solve. Along with the three categories constraints, we operation must be able to perform the operation.
also take armamentarium-constraint into account for mak- According to the three constraints, the operation schedul-
ing the problem more realistic. Besides, we adopt more ing principles were followed, and the special requirements of
granular constraints in “personnel constraints”: a limited emergency operation were added, which met the requirement
number of personnel in different operation stages, restric- of emergency management.
tions on working hours, and flexible overtime hours. The
representative constraints are as follows:
{ } 5 Schedule approach
ETij =STij + max Tijcm ⋅ xijcm − (j − 3)j−1 ri ⋅ Ti1 ,
c∈Cij , m∈Mc
(4) As previously mentioned in Sect. 2, we aim to combine
i ∈ I, j = 1, 2, 3. an advanced machine learning method with the operation
Equation (4) indicates that the end time of a certain stage is scheduling problem equipped with emergency interrup-
the sum of its start time and the maximum operating time of tion and multiple constraints. In this section, we present a
each resource in that stage. However, if it is the first opera- random-forest-based scheduling method with a feedback
tion in the operating room, the preoperative preparation will mechanism in consideration of the complexity of operation
be performed directly in the operating room, so the end time resource scheduling. This system can adjust dispatching
of the operation should be subtracted from the preoperative parameters according to a dynamic operating environment
preparation time. such as changes in the number of patients, available time of
operation resources, the workload of operation resources,
ETij = STi(j+1) , i ∈ I, j = 1, 2, (5) etc. Figure 1 illustrates the framework of the proposed
scheduling method.

13
 L. Li et al.

The proposed scheduling method mainly contains two of operation resources and waiting time of patients, this
research contents: (1) dispatching rules, (2) parameter paper presents CDR through organically combining perti-
optimization model. The “simulation model” is a virtual nent dispatching rules. The composite scheduling rules will
operation resource allocation system constructed based on rank all patients based on their overall priorities (Fig. 2).
a generated data-set. The dispatching rule embedded into The proposed dispatching rules consider the character-
the simulation model is a dispatching policy integrated istics related to process constraint, time constraint, arma-
with patients’ information, operation resources, multiple mentarium constraint, and medical–personnel constraint.
constraints, and some other considerations. Based on the In addition, to better prioritize emergency degrees of
simulation model, we can extract the mass of samples stored operations, the proposed dispatching rules should consider
in the “Samples”. The parameter optimization model is a operation type, estimated operations’ duration, and age of
relational model which can output optimized parameters patients. Figure 3 shows the considerations and the resolving
according to the expected performances. Thus, the schedul- thoughts. A flowchart of the proposed dispatching rules is
ing system can adaptively update dispatching parameters, illustrated in Fig. 4.
form the most suitable dispatching rule, and implement the Step 1 According to information of patients and opera-
adaptive scheduling course. tion resources from clinical department, surgical manage-
ment department, and anesthesiology department, judge
whether the operation is emergent. If the operation is emer-
5.1 Composite dispatching rules gent, turn to Step 2; if not, turn to Step 3.
Step 2 Assign the highest priority to the current emer-
Operation scheduling is a goal-oriented complex combinato- gency operation, compute assigning priority of opera-
rial optimization problem that is similar to the scheduling tion resources according to (13), and arrange operation
problem of flow-shop. Similarly, dispatching rules, aiming immediately.
at rankings all patients in the queue, will greatly affect the Step 3 According to (12), compute urgency degree of
scheduling results. In the existing literature, most of them each selective operation and rank them in descending order
adopted simple heuristic scheduling rules [41–43]. Owing as the operation priority.
to the multiple scheduling targets of the complex operation Step 4 According to (13), compute the assigning priority
scheduling problem, simple heuristic rules are difficult to of each anesthesia room and anesthetist, allocate anesthesia
achieve comprehensive optimization. According to the mul- room and anesthetist with highest priority to the first patient
tiple targets of operation scheduling, such as the utilization in line.

Fig. 2  Framework of the pro-

posed scheduling method

Generate
Samples
pre-processing
Data source

Simulation model

Embed

Parameters optimization model

Generate

Update Dispatching parameters

Dispatching rules

13
Support optimal scheduling with weighted random forest for operation resources

Operation resource dispatching

Research object Affecting factors Methods & theories

Overuse of resources Number of operating rooms Surgery scheduling

Underuse of resources Opening hours of operating rooms Real-time adjustment in operation

Balanced utilization of resources Required equipment Selective or emergency operation

Limited workload of resources Restrictions doctors’ working hours Anesthetist/nurse/doctor scheduling

Queue time of patients Doctors’ scheduling Capacity distribution of resources

Medical costs of patients Number of medical staff Resource-scheduling theory

(α|β|γ)
Utilization of ICU Uncertain operation time α: operating environment
β: characteristics and constraints
... Emergency insertion γ: multi-objective optimization

Fig. 3  Considerations and resolving thoughts of the CDR

Step 5 According to (13), compute the assigning priority Compared with researches on scheduling operation
of each operating room and doctor, allocate operating room resources, which set surgical urgency merely according to
and doctor with highest priority to the first patient in line. the type of operation, CDR in this paper simultaneously con-
Step 6 According to (13), compute the assigning priority sidered operation type, estimated operation time, age (toler-
of each postoperative observation room and nurse, allocate ance) of patients, and utilization of operation resources. The
postoperative observation room and nurse with highest pri- proposed model also assign weight parameters to them to
ority to the first patient in line. facilitate optimization work in the later period.

Ui = 𝛼1 ⋅ Typi + 𝛼2 ⋅ Duri + 𝛼3 ⋅ Agei , (12) 5.2 Parameter optimization based on weighted

random forest
Tijcm
∑
i t − TSw
(13) In academia, there exist plentiful machine learning methods
Pi = + .
TEw − TSw t − ET(i−1)j for mining data potential-ration, such as SVM (support vec-
tor machine), artificial neural network (ANN), ELM (extreme
𝛼1 , 𝛼2 , 𝛼3: weight parameters. learning machine), and RF (random forest). For mining medi-
Typi : operation type of patient i; if the patient’s injury cal knowledge, random forest has been widely used due to its
severity is severe, Typi = 3; if the patient’s injury severity high noise tolerance. For addressing the cancellation prob-
is medium, Typi = 2; if the patient’s injury severity is minor, lem of elective surgeries and improving the performances of
Typi = 1. operation rooms, Li et al. [44] built an RF-based classifica-
Duri : the estimated operation time. According to tion model to identify potential elective surgeries with high
operation type and basic information of patients, Duri is cancellation possibility. For predicting severe postoperative
divided into four levels and assigned the value of 0, 1, 2, 3, complications after cardiac surgeries, Lapp et al. [45] adopted
respectively. and compared several machine learning methods: AdaBoost,
Agei: the age bracket of patients. Patients of different ages gradient boosting model, random forest, and stacking. Their
have different tolerances, this research takes patients’ ages results show that the gradient boosting model and random for-
into account. If the age of the patient is under 5, Agei = 2; est have the best performance with the sensitivity of 0.852
If the age of the patient is over 80, Agei = 1; otherwise, and 0.875, and NPV (negative predictive value) of 0.923 and
Agei = 0. 0.920, respectively. Based on a random forest algorithm, Lin

13
 L. Li et al.

Clinical department Surgical Management department Anesthesiology department

Information of patients and

operation resources

YES Assign the

Emergency operation?
vhighest priority

NO

According to (14), compute urgency degree of operations

According to urgency of operations, rank patients in descending order

According to (15), compute assigning priority of operation resources

(anesthesia room, anesthetist; operating room, doctor; observation room, nurse)

According to (15), assign anesthesia room and anesthetist with higher priority

According to (15), assign operating room and doctor with higher priority

According to (15), assign observation room and nurse with higher priority

Operation schedule

Fig. 4  Framework of the proposed scheduling method

et al. [46] constructed a learning model to predict mortality for second training is conducted after reimporting the first training
acute kidney injury patients in ICUs. The proposed prediction to strengthen the weight of a decision tree with better classi-
model performed better through comparison with two other fication performance, deteriorate the weight of a decision tree
machine learning models and the customized simplified acute with poor classification performance. To improve the prediction
physiology score II model. performance, the error correction strategy is introduced into the
Though random forest has high noise tolerance and generali- prediction model, and the weighted random forest with error
zation performance, to further improve the voting accuracy, this correction is further formed. The process is shown in Fig. 5.
research adds weight adjustment in the conventional random For testing sample x, suppose the output of the lth deci-
forest and forms the weighted random forest: the weight of sion tree is ftree, t (x) = i; i = 1, 2, … , c, i.e., the correspond-
decision trees should be set according to the classification abil- ing category l = 1, 2, … , L; L denotes the number of decision
ity of the constructed decision tree. In addition, to improve the trees in the random forest. Output of the random forest can be
robustness of weights and improve the classification level, the denoted as (14).

13
Support optimal scheduling with weighted random forest for operation resources

W1

Sample set Forecast

W2
Training data Forecasting results
Testing data

Error corredtion
Wn

Forecasting results

Fig. 5  The process of forecasting algorithm

fRF (x) = arg max {I}, the training process. The other part is the testing sample
(14)
i=1,2,…,c
which used to test each decision tree after the training
where I(⋅) represents the number of samples which satisfied process, and to calculate its classification accuracy:
ftree, t (x) = i. Xcorrect, t
The core of the applied weighted random forest model 𝜔l = , l = 1, 2, … , L, (15)
X
is divided into two parts. One part is used as the training
sample of the traditional random forest model to conduct

Fig. 6  Simulation model based on Flexsim 11

13
 L. Li et al.

where Xcorrect, t denotes the number of samples with correct where n denotes the number of samples; Pf (i) and Pr (i),
classification of the lth decision tree and X is the number of respectively, presents the forecast value and actual
testing samples. value.
The correct rate is taken as the weight of the corre- (2) Error correction for predicting results:
sponding decision tree, and each decision tree is multiplied
n Pf (i) � ∑
n
by the weight when voting. Output of the applied weighted 𝛾=
∑
𝜔i ⋅ 𝜔t , (18)
random forest can be expressed in (16). i=1 Pr (i) i=1

where 𝛾 denotes the correction coefficient; 𝜔t presents

� ∑ �
fWRF (x) = arg max 𝜔l . (16)
i=1,2,…,c l∈L, ftree, t (x)=i the coefficient of growth rate.

We adopt coefficient of growth rate to conduct error modi- The final predicting value after error correction can be
fication and the process is as follows: expressed as

(1) Compute the deviation rate: n−i+1

( )
Pf ,c (i) = 1 + ⋅ (1 − 𝛾) ⋅ Pf (i). (19)
n
n �Pf (i) − Pr (i)�
� �
1∑ (17)
𝛽= � � ⋅ 100%,
n i=1 Pr (i)

6 Numerical experiments and analyses

Table 1  Utilization of operation rooms
This section conducts series of numerical experiments to
CDR FIFO CDR FIFO demonstrate the performance of the proposed schedul-
ing method. Section 6.1 details our numerical settings. In
OR_1 76.89 30.14 OR_E_1 86.65 22.18
Sect. 6.2, we make case studies and comparisons between
OR_2 79.55 30.33 OR_E_2 81.07 26.74
the proposed dispatching rules with other FIFO (first in first
OR_3 80.38 18.53 OR_E_3 82.97 43.51
out) to test the advantage of the proposed dispatching rules
OR_4 83.27 36.37 OR_E_4 81.11 21.29
based on the simulation platform. In Sect. 6.3, to test the
OR_5 78.72 25.98 OR_E_5 89.89 40.15
effectiveness of the proposed “feedback mechanism”, we
OR_6 71.6 25.3
conducted a case study and compared proposed dispatch-
OR_7 81.06 29.71
ing rules with the holistic proposed method. Different dis-
OR_8 73.88 26.48
patching parameters will lead to different performances,
so we randomly set parameter values and collected nearly

Table 2  Waiting time of CDR FIFO CDR FIFO CDR FIFO

patients
Patient_1 69 20.42 Patient_17 60.82 24.83 Patient_33 76.23 17.76
Patient_2 76.34 23.37 Patient_18 75.12 15.21 Patient_34 72.34 20.6
Patient_3 76.78 21.13 Patient_19 61.96 19.31 Patient_35 75.52 15.09
Patient_4 79.69 23.82 Patient_20 64.49 21.39 Patient_36 75.15 19.06
Patient_5 69.27 24.18 Patient_21 77.86 15.31 Patient_37 77.87 24.53
Patient_6 78.51 20.33 Patient_22 73.29 19.66 Patient_38 61.45 19.15
Patient_7 62.38 15.08 Patient_23 77.28 20.1 Patient_39 74.41 17.47
Patient_8 69.15 15.39 Patient_24 75.94 16.79 Patient_40 62.68 15.6
Patient_9 63.79 20.29 Patient_25 67.38 21.23 Patient_41 64.36 22.08
Patient_10 73.52 15.31 Patient_26 66.51 22.75 Patient_42 61.98 23.61
Patient_11 68.08 23.05 Patient_27 72.92 17.8 Patient_43 66.99 15.34
Patient_12 66.64 21.07 Patient_28 62.1 16.39 Patient_44 79.12 18.49
Patient_13 75.99 24.9 Patient_29 75.39 16.18 Patient_45 74.71 19.29
Patient_14 65.31 22.23 Patient_30 67.54 20.17 Patient_46 70.4 18.5
Patient_15 76.94 24.94 Patient_31 64.17 17.77 Patient_47 72.92 19.31
Patient_16 77.51 17.38 Patient_32 66.7 21.24 Patient_48 77.51 19.06

13
Support optimal scheduling with weighted random forest for operation resources

20 groups of experimental data to avoid experimental equipment), 36 surgeons, 6 postoperative observation rooms,
coincidence. and 12 nurses. Suppose the hospital works 7 days a week and
8 hours a day, while some operation resources may work
6.1 Experiment description overtime randomly.

To evaluate the performance and limitations of the pro- 6.2 CDR vs. FIFO
posed scheduling method, based on generated basic data of
patients and operation resources, we built simulation model To illustrate the effectiveness of the proposed dispatching
on FlexSim 11 which is shown in Fig. 6. In this experi- rule, in addition to the simulation results of the proposed
ment, we considered 3 anesthesia rooms, 6 anesthetists, 13 dispatching rule, we also recorded the results of FIFO (first
operation rooms (5 of which are equipped with expensive in first out). The simulation results are listed in Tables 1, 2,

Table 3  Operation efficiency of Daily working hours Operation hours Operation hours Efficiency Efficiency
doctors
CDR FIFO CDR FIFO

Doctor_1 8.82 5.9 3.89 0.67 0.44

Doctor_2 8.11 4.82 1.92 0.59 0.24
Doctor_3 8.99 5.93 3.07 0.66 0.34
Doctor_4 8.79 5.98 2.12 0.68 0.24
Doctor_5 8.81 6.75 3.01 0.77 0.34
Doctor_6 8.44 5.2 2.23 0.62 0.26
Doctor_7 8.55 4.85 2.87 0.57 0.34
Doctor_8 8.45 6.47 2.89 0.77 0.34
Doctor_9 8.42 4.34 2.3 0.52 0.27
Doctor_10 8.19 5.75 3.09 0.7 0.38
Doctor_11 8.47 4.87 3.64 0.58 0.43
Doctor_12 8.18 4.78 3.17 0.58 0.39
Doctor_13 8.41 6.39 2.35 0.76 0.28
Doctor_14 8.63 5.43 2.44 0.63 0.28
Doctor_15 8.98 6.24 3.29 0.69 0.37
Doctor_16 8.57 4.03 3.62 0.47 0.42
Doctor_17 8.22 4.98 3.2 0.61 0.39
Doctor_18 8.41 4.06 2.87 0.48 0.34
Doctor_19 8.94 4.87 2.3 0.55 0.26
Doctor_20 8.92 6.73 2.26 0.75 0.25
Doctor_21 8.47 4.18 3.3 0.49 0.39
Doctor_22 8.88 4.7 3.59 0.53 0.4
Doctor_23 8.41 4.68 2.83 0.56 0.34
Doctor_24 8.68 5.9 2.55 0.68 0.29
Doctor_25 8.19 4.04 2.89 0.49 0.35
Doctor_26 8.98 5.81 3.63 0.65 0.4
Doctor_27 8.52 6.24 3.16 0.73 0.37
Doctor_28 8.93 5.16 2.62 0.58 0.29
Doctor_29 8.04 6.04 3.08 0.75 0.38
Doctor_30 8.98 4.38 3.88 0.49 0.43
Doctor_31 8.58 5.05 3.55 0.59 0.41
Doctor_32 8.58 5.63 3.86 0.66 0.45
Doctor_33 8.21 6.34 3.1 0.77 0.38
Doctor_34 8.45 5.5 2.19 0.65 0.26
Doctor_35 8.38 4.89 2.38 0.58 0.28
Doctor_36 8.96 5.18 3.38 0.58 0.38

13
 L. Li et al.

3, where “OR_1” denotes the utilization of NO.1 operation To analyze the simulation results, we carry out normali-
room; “OR_E_1” denotes the utilization of NO.1 operation zation processing on the above simulation results. The nor-
room equipped with expensive equipment. To demonstrate malized results are shown in Fig. 10.
the distinctions between these dispatching rules, we have From Table 4 and Fig. 10, we can draw the following
drawn the line charts presented in Figs. 7, 8, 9. conclusions:
From Tables 1, 2, 3 and Figs. 7, 8, 9, we can draw the
following conclusions: (1) Compared to the dispatching rule, the scheduling
approach improved WT, E_AN, E_SU, E_NU, U_AR,
(1) Compared with FIFO, the utilization of each opera- U_OR_E, U_OR, and U_PO respectively by 8.21%,
tion room has increased by more than 90%, or even 7.59%, 9.79%, 2.30%, 1.64%, 0.94%, 2.21%, and
up to 330%; the waiting time of each patient has been 1.23%. This can illustrate the usefulness and effective-
reduced by more than 180%, or up to 400%; the opera- ness of “feedback-mechanism”.
tion efficiency of each doctor has increased by more (2) Regardless of the dispatching parameters, the proposed
than 11%, even reaching 198%. scheduling approach can effectively improve perfor-
(2) The utilization of each operation room could maintain mances and remain within a relatively stable range.
relative stability: fall between 70 and 90%; the waiting (3) The proposed scheduling approach could respond to the
time of each patient can be kept relative stability: fall changing manufacturing environment, such as sudden
in between 60 and 80 hours. addition of emergency operation, some uncertain factor
(3) Owing to the various influence factors such as sur- of operation resources and patients.
geons’ capacity, scheduling, and random vacating, the (4) From these above results, we can confirm that the pro-
stability of doctors’ operation efficiency is not good. posed method works to maximize the utilization of
(4) The high utilization of operation resources came from operation resources and minimize the waiting time of
high response to all patients under scheduling plus the patients.
cases of accident-related incidents.

6.3 SSA vs. CDR

The proposed SSA is validated through simulation experi- 90.00

80.00
ments and comparison with the CDR. Different dispatching
70.00
parameters could result in different performances, so we 60.00
have made dozens of experiments with different dispatch- 50.00
ing parameters to avoid contingency. The simulation results 40.00
30.00
are shown in Table 4. The dispatching rule refers to the 20.00
proposed dispatching rule introduced in Sect. 5.1 (without 10.00
feedback mechanism). The scheduling approach refers to the 0.00
P_1
P_3
P_5
P_7
P_9
P_11
P_13
P_15
P_17
P_19
P_21
P_23
P_25
P_27
P_29
P_31
P_33
P_35
P_37
P_39
P_41
P_43
P_45
P_47
proposed SSA (the proposed dispatching rule & feedback
mechanism). CDR FIFO

Fig. 8  Waiting time of patients

100.00 0.90
90.00 0.80
80.00 0.70
70.00
60.00 0.60
50.00 0.50
40.00 0.40
30.00 0.30
20.00 0.20
10.00
0.00 0.10
0.00
O _1

O _2

O _3

O _4
_5
_1

_2

_3

_4

_5

_6

_7

_8
_E

_E

_E

_E

_E

_1
_3
_5
_7

D 9
D 1
D 3
D 5
D 7
D 9
D 1
D 3
D 5
D 7
D 9
D 1
D 3
5
O

_
_1
_1
_1
_1
_1
_2
_2
_2
_2
_2
_3
_3
_3
R

D
D
D
D
D
O

CDR FIFO CDR FIFO

Fig. 7  Utilization of operation rooms Fig. 9  Operation efficiency of doctors

13
Support optimal scheduling with weighted random forest for operation resources

Table 4  Simulation results of the SSA

NO. Waiting time Efficiency Efficiency surgeon Efficiency nurse Utilization AR Utilization OR_E Utilization OR Utilization PO
anesthetist

1 67.36 71.42 70.41 83.7 82.39 84.11 80.74 85.92

2 67.44 73.46 66.07 83.37 82.67 84.35 81.98 85.4
3 62.53 73.7 69.87 85.79 84.81 86.31 79.56 86.48
4 61.54 72.77 69.6 85.8 84.51 85.73 81.17 85.75
5 65.05 71.81 71.38 85.64 83.73 84.99 78.78 87.23
6 65.48 72.86 71.2 84.96 84.04 85.22 79.45 85.1
7 66.06 72.42 67.31 83.45 83.35 84.3 79.81 87.95
8 62.47 74.05 70.61 85.84 83.18 85.36 79.21 86.56
9 66.46 71.82 67.85 86.04 84.61 84.8 78 86.1
10 65.97 72.74 71.52 86.99 82.22 84.44 78.75 87.94
11 63.78 72.65 68.65 85.65 82.92 85.01 82.12 87.13
12 61.46 74.94 66.13 83.51 82.07 86.35 79.02 85.54
13 65.27 74.87 67.45 84.67 83.6 86.83 80.41 85.19
14 66.41 71.74 71.7 86.95 84.06 84.27 79.32 85.32
15 67.09 74.54 69.07 82.45 84.81 84.48 79.25 87.46
16 60.29 71.75 66.62 83.83 83.51 85.55 80.73 85.2
Ave. 64.67 72.97 69.09 84.92 83.53 85.13 79.89 86.27
CDR 70.45 67.82 62.93 83.01 82.18 84.34 78.17 85.22
Imp. 8.21% 7.59% 9.79% 2.30% 1.64% 0.94% 2.21% 1.23%

characteristics of operating resources were combined to

1.10 make the systematic research more consistent with the actual
1.08 situation. CDR, optimization ideology, and feedback mecha-
1.06
nism were integrated into this research, which could realize
1.04
1.02 closed-loop optimizations adaptively based on random forest
1.00 algorithm combined with some resources’ information such
0.98 as flexible operating time, hold-up time of medical facili-
0.96 ties, available time of medical staff, and multiple constraints.
0.94
Average Dispatching rule Experiments on a simulation platform were carried out to
verify the effectiveness of the proposed scheduling method.
WT E_AN E_SU E_NU
Through the discussion of operating resources schedul-
U_AR U_OR_E U_OR U_PO
ing theory and methods, some of its new future research
directions can be considered: strengthening the combina-
Fig. 10  Comparison of SSA with dispatching rules
tion of production scheduling theory and the actual situ-
ation of operations, and improving the basic framework
Concisely, different dispatching parameters could result in and formal research issues of operating resources sched-
different performances. Therefore, we could improve perfor- uling. The various models in scheduling are worthy of
mances by changing dispatching parameters appropriately. discussion, and the influence of different uncertainties in
However, the operation status is ever-changing, we cannot operation should also be considered when applying sched-
confirm dispatching parameters which suitable for all pro- uling methods in the future. In the future work, a more
duction statuses. It is possible if we make the scheduling reasonable mathematical model for dynamic operating
system be capable of adapting the dispatching parameters. room scheduling needs to be established and an efficient
algorithm should be developed to solve the emergency
resource scheduling problem under emergencies.
7 Conclusions
Acknowledgements This research was supported by the National
In this paper, resource scheduling in hospital operating Key R&D Program of China (No. 2018YFE0105000), the Shanghai
Municipal Commission of Science and Technology (No. 19511132100)
rooms was studied and analyzed. In the process of model and the National Natural Science Foundation of China (No. 51475334).
construction, scheduling theory, queuing idea, and multiple

13
 L. Li et al.

References 19. Cardoen, B., Demeulemeester, E., & Beliën, J. (2010). Operat-
ing room planning and scheduling: a literature review. European
Journal of Operational Research, 201(3), 921–932.
1. Pinker, E., & Tezcan, T. (2013). Determining the optimal con-
20. Pham, D. N., & Klinkert, A. (2008). Surgical case scheduling as
guration of hospital inpatient rooms in the presence of isolation
a generalized job shop scheduling problem. European Journal of
patients. Operations Research, 61(6), 1259–1276.
Operational Research, 185(3), 1011–1025.
2. Cardoso, T., Oliveira, M. D., Barbosa-Póvoa, A., & Nickel, S.
21. Dexter, F., Wachtel, R. E., & Epstein, R. H. (2016). Decreasing
(2016). Moving towards an equitable long-term care network:
the hours that anesthesiologists and nurse anesthetists work late
a multi-objective and multi-period planning approach. Omega,
by making decisions to reduce the hours of over-utilized operating
58, 69–85.
room time. Anesthesia and Analgesia, 122(3), 831–842.
3. Cardoso, T., Oliveira, M. D., Barbosa-Póvoa, A., & Nickel, S.
22. Lamiri, M., Xie, X. L., Dolgui, A., & Grimaud, F. (2008). A sto-
(2015). An integrated approach for planning a long-term care
chastic model for operating room planning with elective and emer-
network with uncertainty, strategic policy and equity consid-
gency demand for operation. European Journal of Operational
erations. European Journal of Operational Research, 247(1),
Research, 185(3), 1026–1037.
321–334.
23. Wachtel, R. E., & Dexter, F. (2008). Tactical increases in operat-
4. Xiang, W., Yin, J., & Lim, G. (2015). An ant colony optimiza-
ing room block time for capacity planning should not be based on
tion approach for solving an operating room operation scheduling
utilization. Anesthesia and Analgesia, 106(1), 215–226.
problem. Computers and Industrial Engineering, 85, 335–345.
24. Hans, E., Wullink, G., Houdenhoven, M. V., & Kazemier, G.
5. Silva, T., Souza, M. C., Saldanha, R. R., & Burke, E. K. (2015).
(2006). Robust surgery loading. European Journal of Operational
Surgical scheduling with simultaneous employment of specialised
Research, 185(3), 1038–1050.
human resources. European Journal of Operational Research,
25. Shu, W., & Luo, L. (2008). A surgical scheduling study based
245(3), 719–730.
on objective programming. Technology and Market, 2, 42–44 (in
6. Maenhout, B., & Vanhoucke, M. (2013). An integrated nurse
Chinese).
stafng and scheduling analysis for longer-term nursing staff allo-
26. Zhong, L. W., Luo, S. C., Yang, G. P., Wang, G. Y., Xu, L., Zhao,
cation problems. Omega, 41(2), 485–499.
C. M., & Zhang, P. (2009). Study on surgery scheduling optimiza-
7. Dios, M., Molina-Pariente, J. M., Fernandez-Viagas, V., Andrade-
tion and computer program management in operating room. Jour-
Pineda, J. L., & Framinan, J. M. (2015). A decision support sys-
nal of Shanghai Second University of Technology, 26(4), 286–290
tem for operating room scheduling. Computers and Industrial
(in Chinese).
Engineering, 88, 430–443.
27. Bai, X., Luo, L., & Li, M. R. (2011). The present situation and
8. Zhang, L. M., Chang, H. Y., & Xu, R. T. (2013). The patient
development prospect of the study on surgical scheduling in
admission scheduling of an ophthalmic hospital using genetic
hospital management. Management Review, 23(1), 121–128 (in
algorithm. Advanced Materials Research, 756, 1423–1432.
Chinese).
9. Lu, Y., Xie, X., & Jiang, Z. (2018). Dynamic appointment sched-
28. Kamran, M. A., Karimi, B., & Dellaert, N. (2018). Uncertainty in
uling with wait-dependent abandonment. European Journal of
advance scheduling problem in operating room planning. Comput-
Operational Research, 265(3), 975–984.
ers and Industrial Engineering, 126, 252–268.
10. Andersen, A. R., Nielsen, B. F., & Reinhardt, L. B. (2017).
29. Xiao, G., Jaarsveld, W., Dong, M., & Klundert, J. (2016). Stochas-
Optimization of hospital ward resources with patient relocation
tic programming analysis and solutions to schedule overcrowded
using Markov chain modeling. European Journal of Operational
operating rooms in China. Computers and Operations Research,
Research, 260(3), 1152–1163.
74, 78–91.
11. Qi, J. (2016). Mitigating delays and unfairness in appointment
30. Pang, B., Xie, X., Song, Y., & Luo, L. (2019). Operation schedul-
systems. Management Science, 63(2), 566–583.
ing under case cancellation and operation duration uncertainty.
12. Guo, H., Gao, S., Tsui, K. L., & Niu, T. (2017). Simulation opti-
IEEE Transactions on Automation Science and Engineering,
mization for medical staff conguration at emergency department
16(1), 74–86.
in Hong Kong. IEEE Transactions on Automation Science and
31. Molina-Pariente, J. M., Fernandez-Viagas, V., & Framinan, J.
Engineering, 14(4), 1655–1665.
M. (2015). Integrated operating room planning and scheduling
13. Zhong, X., Li, J., Bain, P. A., & Musa, A. J. (2017). Electronic
problem with assistant surgeon dependent operation durations.
visits in primary care: modeling, analysis, and scheduling poli-
Computers and Industrial Engineering, 82, 8–20.
cies. IEEE Transactions on Automation Science and Engineering,
32. Erhard, M., Schoenfelder, J., Fügener, A., & Brunner, J. O. (2018).
14(3), 1451–1466.
State of the art in physician scheduling. European Journal of
14. Ahmadi-Javid, A., Jalali, Z., & Klassen, K. J. (2017). Outpatient
Operational Research, 265(1), 1–18.
appointment systems in healthcare: a review of optimization stud-
33. Qiu, C., Cannesson, M., Morkos, A., Nguyen, V. T., LaPlace, D.,
ies. European Journal of Operational Research, 258(1), 3–34.
Trivedi, N. S., et al. (2016). Practice and outcomes of the perio-
15. Huh, W. T., Liu, N., & Truong, V. A. (2013). Multiresource allo-
perative surgical home in a california integrated delivery system.
cation scheduling in dynamic environments. Manufacturing and
Anesthesia and Analgesia, 123(3), 597–606.
Service Operations Management, 15(2), 280–291.
34. Zhou, L., Geng, N., Jiang, Z., & Wang, X. (2017). Combining
16. Liu, N., Truong, V. A., Wang, X., & Anderson, B. (2019). Inte-
revenue and equity in capacity allocation of imaging facilities.
grated scheduling and capacity planning with considerations for
European Journal of Operational Research, 256(2), 619–628.
patients’ length-of-stays. Production and Operations Manage-
35. Zhou, L., Geng, N., Jiang, Z., & Wang, X. (2018). Multi-objec-
ment, 28(7), 1735–1756.
tive capacity allocation of hospital wards combining revenue and
17. Addis, B., Carello, G., Grosso, A., & Tánfani, E. (2016). Operat-
equity. Omega, 81, 220–233.
ing room scheduling and rescheduling: a rolling horizon approach.
36. Kim, K., & Mehrotra, S. (2015). A two-stage stochastic integer
Flexible Services and Manufacturing Journal, 28(1–2), 206–232.
programming approach to integrated stafng and scheduling with
18. Magerlein, J. M., & Martin, J. B. (1978). Surgical demand sched-
application to nurse management. Operation Researches, 63(6),
uling: a review. Health Services Research, 13(4), 418–433.
1431–1451.

13
Support optimal scheduling with weighted random forest for operation resources

37. Rath, S., Rajaram, K., & Mahajan, A. (2017). Integrated anesthe- unmanned systems. She has over 80 publications including 4 books,
siologist and room scheduling for surgeries: Methodology and 50+ journal papers, and 2 book chapters.
application. Operation Researches, 65(6), 1460–1478.
38. Ali, H. H., Lamsali, H., & Othman, S. N. (2019). Operating rooms
scheduling for elective surgeries in a hospital affected by war- Qingyun Yu is a Ph.D. candidate
related incidents. Journal of Medical Systems. https://​doi.​org/​10.​ of control science and engineer-
1007/​s42417-​020-​00265-8. ing at Tongji University. She
39. Al, R. A., Chen, T., & Judeh, M. (2018). Optimal operating received her B.Sc. degree in
room scheduling for normal and unexpected events in a smart automation from Jiangnan Uni-
hospital. Operation Researches, 18(3), 579–602. versity, China in 2013. Her
40. Park, H. S., Kim, S. H., Bong, M. R., Choi, D. K., Kim, W. J., research interests are in produc-
Ku, S. W., et al. (2020). Optimization of the operating room tion planning and scheduling,
scheduling process for improving efficiency in a tertiary hospi- computational intelligence and
tal. Journal of Medical Systems, 44(171), 1–7. semiconductor manufacturing.
41. Assad, D. B. N., & Spiegel, T. (2019). Maximizing the effi-
ciency of residents operating room scheduling: a case study at
a teaching hospital. Production, 29, e20190025.
42. Zhou, L. P., Geng, N., Jiang, Z. B., & Wang, X. X. (2020).
Public hospital inpatient room allocation and patient scheduling
considering equity. IEEE Transactions on Automation Science
and Engineering, 17(3), 1124–1139. Haoyi Shi is an M.Sc. candidate
43. Gartnera, D., & Padmanb, R. (2019). Flexible hospital-wide of control science and engineer-
elective patient scheduling. Journal of The Operational ing at Tongji University. He
Research Society, 71(6), 878–892. received his B.Sc. degree in
44. Li, L., Liu, C., Li, F., Zhao, S. Z., & Gong, R. R. (2018). A automation from Tongji Univer-
random forest and simulation approach for scheduling opera- sity, China in 2019. His research
tion rooms: Elective surgery cancelation in a Chinese hospital interest is in machine learning.
urology department. International Journal of Health Planning
and Management, 33(4), 941–966.
45. Lapp, L., Bouamrane, M. M., & Kavanagh, K. (2019). Evalua-
tion of random forest and ensemble methods at predicting com-
plications following cardiac surgery. Artificial Intelligence in
Medicine, 11526, 376–385.
46. Lin, K., Hu, Y. H., & Kong, G. L. (2019). Predicting in-hospital
mortality of patients with acute kidney injury in the ICU using
random forest model. International Journal of Medical Infor-
matics, 125, 53–61.
Yuguang Liu is an M.Sc. candi-
date of control science and engi-
neering at Tongji University. He
Li Li is Professor of control sci-
received his B.Sc. degree in
ence and engineering at Tongji
automation from Tongji Univer-
University. She received her
sity, China in 2019. His research
B.Sc. degree and M.Sc. degree
interest is in machine learning.
in electrical automation from
Shenyang Agriculture Univer-
sity, China, in 1996 and 1999,
respectively, and Ph.D. degree in
mechatronics engineering from
Shenyang Institute of Automa-
tion, Chinese Academy of Sci-
ence in 2003. Her research inter-
ests are in intelligent
manufacturing, machine learn-
ing, artificial intelligence, and

13