Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)

IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

Water Quality Analysis and Prediction using

Machine Learning
Dr. D. Brindha Viswanath Puli Bala Karthik Sobula NVSS
dept. of Computer Science and dept. of Computer Science and dept. of Computer Science and
Engineering Engineering Engineering
(Assistant Professor) (Bachelor’s of Technology) (Bachelor’s of Technology)
Karunya Institute of Technology and Karunya Institute of Technology and Karunya Institute of Technology and
2023 7th International Conference on Computing Methodologies and Communication (ICCMC) | 978-1-6654-6408-6/23/$31.00 ©2023 IEEE | DOI: 10.1109/ICCMC56507.2023.10083776

Sciences Sciences Sciences

(Deemed to be University) (Deemed to be University) (Deemed to be University)
Coimbatore, India Coimbatore, India Coimbatore, India
[email protected] [email protected] [email protected]

Vamsi Stephen Guru Dinesh

M ittakandala Nanneboina
dept. of Computer Science and dept. of Computer Science and
Engineering Engineering
(Bachelor’s of Technology) (Bachelor’s of Technology)
Karunya Institute of Technology and Karunya Institute of Technology and
Sciences Sciences
(Deemed to be University) (Deemed to be University)
Coimbatore, India Coimbatore, India
[email protected] [email protected]

Abstract — The main objective of this research is to estimate In the current scenario, determining water quality is a
the water quality using machine learning technique. Water is time-consuming process that involves lots of statistical
considered as a vital resource that has an impact on many facets analysis and is quite expensive, which also requires a lot of
of human health and existence. People who live in metropolitan time and effort to calculate. The main objective of this work
areas are often concerned about the quality of the water as it is is to provide the quality score of water and provide the
critical to monitor the quality of water. Water sample collection suggestion if the water needs to treated/purified or not, based
and laboratory analysis are time and resource-intensive on the water quality score that is calculated by the machine
processes. Analyzing water quality is a complicated subject learning model.
because of the many variables that affect it. This concept is
inextricably linked to the various purposes for which water is II. LITERATURE SURVEY
used. The goal of this study is to estimate water quality by
acquiring several parameters, and using the machine learning
method, Random Forest regression. In this case, the model uses The methods used to address problems with water quality
parameters like pH, turbidity, dissolved oxygen, conductivity, and have been reviewed. In previous research, tradit ional lab
others. analysis and statistical analysis are frequently employed to
help determine the quality of the water, while other analys is
Keywords—Water Quality, Machine Learning, Random
use machine learn ing approaches to help identify the best
Forest regression, Decision Tree, Web User Interface (UI),
possible solution to the water quality problem.
I. INT RODUCT ION
In [1], Osim Ku mar Pal et al., proposed that predicting
Water is the most crucial supply and resource for
potable quality of water is more beneficial for water
maintaining a wide range of life. Rapid industrialization distribution and environmental protection. Contaminated
therefore resulted in the deterioration of quality of water at
water causes substantial waterborne infections and presents a
an astonishing pace. The foremost reason for the spread of hazard to human health. Estimating the quality of
terrible illnesses is recognized to be poor water quality. consumable water, may lower the frequency of water-related
Approximately 2 b illion people worldwide consume water
disorders. The latest method for machine learn ing has
that has been contaminated with faeces . Drinking such demonstrated promising accurate performance for water
contaminated water puts humans at risk for water-borne
quality. The study utilized five distinct learning algorith ms to
illnesses. According to the research by WaterAid India, one estimate drinking water quality. Firs t, data is acquired fro m
in ten individuals do not have access to safe drinking water.
several public sources and reported in compliance with the
Water sources are polluted more and more due to over use. water quality criteria of World Health Organizat ion (WHO).
The key issue may be the limited purification methodology
Artificial Neural Network (ANN) provided 99 % highest
that ensures the water quality. According to the report given
accuracy with a 0.75% training error during the training
by UNICEF, it is estimated that the water-borne ailments phase. Random Forest achieved an 87.76% F1 score and
cost India's economy over 600 million USD per year.
82.45% prediction accuracy. The ANN predicted high with
96.51% of F1 score.

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 175

Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on December 01,2024 at 13:27:12 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

In [2], A mir Hamzeh Haghibi et al., proposed a AI-based III. PROPOSED A PPROACH A ND M ET HODOLOGY
framework for water quality expectations. The study
investigated the presentation of AI approaches, such as A. Methodology Flow Diagram
ANN, group information handling strategies, and Support
Vector Machines (SVM) for predicting quality of water
portions of the River Tireh in Iran. Various types of move
and piece capabilities were tested separately to build up the
ANN and SVM . Analyzing the results of both, it showed that
the two models had a good ability to predict water quality
segments. Seven parameters such as, pH, SO4 , Na, Ca, Cl,
Mg, and HCO3 were considered.
In [3], Salisu Yusuf Muhammad et al. used Machine
Learning Techniques to implement a classification model for
finding the quality of water. Based on AI computations, the
article presented an acceptable grouping approach for sorting
water quality. The study looked into and studied the
representation of many arrangement models and calculations
to determine the main aspects that contributed to managing
the water nature of the River Kinta located in Malaysia. The
5 separate models with accurate calculations were assessed,
contrasted, and demonstrated. In examin ing the outcomes, K-
Star Calculation which is used from the Lazy Model, was the Fig. 1. Flow Diagram of Methodology
most effective with 86.67% preciseness. In general,
wastewater is dangerous to human lives, and developing B. Data Collection
rational models to solve this problem is necessary.
The dataset collected from Kaggle consists of 1992
In [4], Consolata Gakii et al., presented a water quality samples of various Indian rivers from the years 2004 to 2014.
analysis classification model based on Decision Tree. The The dataset consists of several parameters of river data such
study proposed a characterization methodology that utilizes a as year, Dissolved Oxygen, Temperature, pH values,
decision tree to monitor the quality of water data fro m Conductivity, Biochemical Oxygen Demand (B.O.D),
several areas in East African country Kenya. The quality of Nitrate Nitrogen, and Fecal Coliforms.
water is critical in ensuring that inhabitants consume water
which is safe to drin k. The usage of decision tree as an For predicting the water quality of the user entered data,
informational examination approach to predict the water is the trained model is used. The data is accepted through a web
drinkable water or not based on water quality parameters can application. After calculation of the water quality score, the
assist the research institute technician by predicting which predicted score is displayed on the web application with a
test results of water should proceed to the next stage of the suggestion of what measure to be taken on the water before
inquiry. The Kenya Water Institution's optional information consuming it.
was used in the building of the model. WEKA programming
was used to run the information model. To order/predict the
clean and unclean water, grouping using a decision tree was
used. The alkalinity water, pH level of water, and
conductivity tests may all play a part in determining quality
of water. The five classifiers of decision tree i.e., J48
algorithm, Logistic Model Tree, Random Timberland, and
Decision Stump, Hoeffd ing tree, were used to build the
model and assess its exactness. J48 decision tree achieved
94% maximu m accurate result, whereas Decision Stu mp
attained 83% which was the poorest precision.
In [5], C.Ashwini et al. presented a paper using Internet
of Things (IoT) and Machine Learn ing to monitor the quality
of water. The paper provided a practical solution to prevent
private overhead tanks' water fro m being contaminated. IoT Fig. 2. Interface for accepting user inputs
devices were used to observe the characteristics of water, and
AI computations were used to predict future water C. Data Preprocessing
contamination. To obtain the required water parameters, the The data collected from Kaggle consists of null values
proposed structure included several types of sensors and duplicate values. Therefore, before analyzing the data,
connected to NodeMCU module. The structure aids in the the data needs to be cleaned. After the data cleaning, the
conservation of water and is also cost effective. The values are normalized to a 0 to 100 range utilizing q-values,
following seven parameters, pH value of water, temperature so that the Water Quality Index (WQI) can be calculated
of water, turbidity, color of water, dissolved oxygen, total utilizing the available parameters . After the calculation of
natural carbon and conductivity were considered and WQI, all actual data values are normalized using z-score. As
evaluated. a result, the values are on the same scale. Following that is
the entire procedure.

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 176

Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on December 01,2024 at 13:27:12 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

TABLE 1 Standard limits of parameters

S.No
. Parameter Standard Limit
1 Dissolved Oxygen 5.0 to 6.0 mg/lit

2 Chloride 250 mg/lit

3 pH 6.5 to 8.5
4 Chemical Oxygen Demand 250 mg/lit
5 Conductivity 0.1 to 1.5 mS/cm
Fig.3.2. Univariate Analysis of normalized parameters
6 Nitrate-Nitrogen 45 mg/lit —Box plot
7 Alkalinity 200 mg/lit

8 Fecal Coliforms zero

9 T otal Dissolved Solids 500 mg/lit

10 Sulphates 200 mg/lit

11 Fluorides 0.6 to 1.2 mg/lit

D. Data Analysis
1) Normalization Fig.3.3. Univariate Analysis on WQI of each parameter
—Box plot
To normalize the values, specifically the water quality
parameters, Q-value normalization is employed for fitting
the parameters in the range of values 0 to 100 for simpler
indexed computation. Every parameter that is used in the
model is normalized.

Where, Vn is the value of normalization,

Vmax is the maximum value of the parameter,
Vmin is the minimum value of the parameter,
V is Vmin (if V value is minimum, then V=0),
V is Vmax (if V value is maximum, then V=1).

The most basic form of statistical data analysis

technique is univariate analysis. The data in this
Fig.3.4. Univariate Analysis—Dist. plot
situation contains just one variable and does not include
a cause-and-effect relationship. 2) Bivariate Analysis
Bivaritae analysis includes the investigation of t wo
parameters, with the objective of identifying the
emp irical correlation between them. Bivariate analysis
can be useful in evaluating basic hypotheses of
relationship. The below figures are relationships of WQI
with parameters.

Fig.3.1a-3.3.1h. Univariate Analysis of parameters—Box plot

Fig.3.5. Multivariate Analysis

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 177

Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on December 01,2024 at 13:27:12 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

It is the statistical analysis of data in which challenge, with individual leaf hub comparing to a class
numerous measurements are taken on each experimental mark and parameters represented on the inner core of
unit and the correlations between multivariate tree. Decision tree can deal with challenges involving
measurements and their structure are crucial. both literal and mathemat ical data. Choice Tree topology
generates a tree like structure where each inner hub
Heat map- The use of colors to signify the value of
symbolizes a test quality, individual branch symbolizes
each data point in a heat map visualization is a way of the test's unintended result, and every leaf core provides
visually expressing numerical data. The warm-to-cool
the sensible option.
color scheme is most widely utilized in heat map
visualization, with warm colors indicating high-value 2) Random Forest Regression
data points and cold colors representing low-value data
points. A random forest is a meta-estimator that uses
averaging to improve projected accuracy and avoid over-
fitting by fitting a range of classification decision trees on
distinct sub-samples of the dataset.

G. Data Splitting
The final step for applying the Machine Learning
algorithm is data splitting. After applying the machine
leaning model, the model is tested with a particular
Fig.3.6a and 3.6b. Multivariate Analysis—Heat map portion of the data to evaluate performance of the model
that is trained and compute the accuracy measurements.
3) Correlation Analysis
In fig. 5, the data split is done for train ing and
To locate the variables that are dependent and to testing in which 80% of data is used for training and 20%
estimate tough variables using readily achievable for testing. After splitting the data, the data is fitted into a
parameters, the correlation analysis is used to uncover random forest regression algorithm.
the feasible correlations between parameters.

Fig. 5. 1 Data splitting

Fig.3.7. Correlation Analysis of the values

Fig. 5. 2 Fitting the data in random forest regression
E. Water Quality Index
H. Evaluation Metrics
Water quality index is a single metric that shows quality
score of water and is derived utilizing numerous criteria that 1) Mean Square Error
are actually indicative of water quality. To co mpute the WQI Mean Squared Error (MSE) is the average of the
conventionally, six water quality parameters are used fro m squared difference between the expected and original
the dataset. Considering these variables and their values in a data collection. MSE calculates the error
corresponding weightages, the WQI calculation is done on terms' variance.
each data as represented in the below mentioned formula, in
which q value represents the values of parameters in the range MSE = ∑
of 0–100 and wfactor reflects the weight of a specific
parameter. Where, N is the number of samples,
∑ is the predicted value,
WQI = is the actual value.
∑

2) Root Mean Square Error

Root Mean Squared Error (RMSE) is the square
root of Mean Squared Error. This parameter co mputes
the standard deviation of the residuals.

Fig. 4. Univariate and Bivariate Analysis of WQI

RMSE = √ =√ ∑
F. Machine Learning Algorithms
1) Decision Tree: 3) Mean Absolute Error

Managed learning is the category that describes the The Mean Absolute Error (MA E) is the average of
estimate of a decision tree. Both agreement and relapse the absolute difference between the dataset's actually
issues may be addressed with their help. The decision present and predicted values. The dataset's residuals are
tree employs a tree representation to address the averaged out in this measurement.

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 178

Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on December 01,2024 at 13:27:12 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

MAE = ∑ lot of study being done on predicting water quality.

Normally, a co llect ion of chemical factors and physical
4) R—Squared factors that are closely connected to the water's intended use
The coefficient of determination, often known as must be used to determine water quality. Then, acceptable
R2 , is the proportion of the dependent variable's values and unacceptable values for each variable should be
fluctuation that the linear regression model can explain. established. Water is deemed acceptable for an application
Because the score is not scaled, it is always less than when its variables co mply with the requirements set out in
one, regardless of how large or small the numbers are. advance. The water must be treated before usage if it doesn't
match these requirements. Nu merous physical and chemical
∑
R2 = traits may be utilized to assess the level of water pollution or
∑ the quality of the water. As a result, it is not practical to
IV. RESULT S describe quality of water precisely, either on a geographical
or temporal basis, by analyzing the behavior of each
The methodology and idea used in implementing this individual variable independently. The challenging option is
work provides a high accuracy model, and the model will
to combine the values of a number of physical factors and
also accept user inputs using the web interface. The R-
chemical factors to create a s ingle value, such as an overall
squared value obtained fro m Random Forest Regression
model is compared with Linear Regression model. After index. The water quality index measures the effectiveness of
accepting the user inputs, they are processed through the any water, that may be created by only identify ing the
model and the score is estimated. Based on the score, it requirements needed for that usage. This indicator has a
predicts how safe the water is and also tells how intense number of chemical and physical properties. Each variab le
should the treatment or purification be. As this work is in the index has a quality value function that indicates how
based on Machine Learning algorithm, Random Forest well the parameter and its quality level are equivalent.
Regression Squared error has high weightage after Direct measurements of a substance's or an entity's
training and testing the model, so it predicts the score concentration are used to define a physical variable's value
utilizing squared error. as determined by water samp le tests. The primary goal o f
this research to analyze mach ine learning methods for
predicting the quality of water using eight different factors,
including temperature, conductivity, pH, d issolved oxygen
(% sat), nit rates, and fecal and total coliforms, has been
achieved.
Fig. 6. 1 Performance Metrics of the Random Forest Regression
Model VI. FUT URE WORK
In future, a lot of improvements can be done in
terms of upgrading the model’s accuracy and pred iction.
The data which is entered by user could be stored in
database, so that it can be used for future research analysis.
This model requires six parameters to predict the water
quality, and in future, the parameters can be reduced and can
Fig. 6. 2 R2 Score on Training Data and Testing Data of Random make it availab le for all users. More functionalities and
forest regression stylings can be added to the web interface, so that the web
application can be more interactive. The web application is
currently limited to the users who have knowledge on the
water quality parameters that are used in this work; in
future, this problem could be overco me, and all users can
Fig. 6. 3 R2 Score on Testing data of Linear Regression use the web application to know the quality of water. Fo r
now, the model predicts the water quality with few
suggestions for the purification and treat ment of water;
however, in the future work, the suggestions can be
elaborated with what minerals can be added to the water.

REFERENCES
[1] Osim Kumar Pal (2022) “The Quality of Drinkable Water using
Machine Learning T echniques” – International Journal of Advanced
Engineering Research and Science
Fig. 6. 4 UI after calculating results
[2] Amir Hamzeh Haghibi, Ali Heidar Nasrolahi, Abbas Parsaie (2018),
“Water quality prediction using machine learning”– Journal of Water
V. CONCLUSION quality research.
The examination of water quality is a co mplex issue because [3] Salisu Yusuf Muhammad, Mokhairi Makhtar, Azilawati Rozaimee,
of the many variables that affect it. Th is idea, in part icular, Azwa Abdul Aziz and Azrul Amri Jamal (2015), “Classification
Model for Water Quality using Machine Learning T echniques” -
is inextricably lin ked to the many purposes for which water International Journal of Software Engineering and Its Applications
is used. Different needs call for various standards. There is a Vol. 9, No. 6 , pp. 45-52

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 179

Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on December 01,2024 at 13:27:12 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the 7th International Conference on Computing Methodologies and Communication (ICCMC-2023)
IEEE Xplore Part Number: CFP23K25-ART; ISBN: 978-1-6654-6408-6

[4] Consolata Gakii and Jeniffer Jepkoech (2019), “ A Classification [14] Mengyuan Zhu, Jiawei Wang, Xiao Yang Yu Zhang, Linyu Zhang,
Model for Water Quality Analysis Using Decision Tree”- European Hongqiang Ren, Bing Wu, LinYe (2022). “A review of the
Journal of Computer Science and Information Technology Vol.7, application of machine learning in water quality evaluation”. - Eco-
No.3, pp.1-8. Environment & Health Vol 1 Issue 2 pp. 107-116
[5] C.Ashwini, Uday Pratap Singh, Ekta Pawar, Shristi (2019), “Water [15] Jui-Sheng Chou, Chia-Chun Ho, Ha-Son Hoang (2018). “Determining
Quality Monitoring Using Machine Learning And IOT” - quality of water in reservoir using machine learning”. Ecological
International journal of scientific & technology research volume 8, Informatics Vol 44 pp. 57-75.
issue 10. [16] Mourade Azrour, Jamal Mabrouki, Ghizlane Fattah, Azedine Guezzaz
[6] J. I. Ubah, L. C. Orakwe , K. N. Ogbu , J. I.Awu , I. E.Ahaneku & E. & Faissal Aziz (2021). “Machine learning algorithms for efficient
C. Chukwuma (2021), “Forecasting water quality parameters using water quality prediction”. Modeling Earth Systems and Environment
artifcial neural network for irrigation purposes” – Vol 8, pp 2793–2801
Nature.com/scientific-reports [17] Vesna Ranković, Jasna Radulovic, Ivana D. Radojevic, Aleksandar
[7] Nida Nasira, Afreen Kansal, Omar Alshaltone, Feras Barneih, Ostojić, Ljiljana Čomić(2010) .” Neural network modeling of
Mustafa Sameer, Abdallah Shanableh, Ahmed Al-Shamma'a (2022). dissolved oxygen in the Gruža reservoir, Serbia”-- Ecological
“Water quality classification using machine learning algorithms”-- Modelling Vol 221(8) pp. 1239-1244
Journal of Water Process Engineering [18] Rajiv Das Kangabam, Sarojini Devi Bhoominathan, Suganthi
[8] Zhao Fu (2020), “Water Quality Prediction Based on Machine Kanagaraj & Munisamy Govindaraju(2017). “Development of a water
Learning T echniques”-- UNLV T HESES, DISSERTAT IONS, quality index (WQI) for the Loktak Lake in India” -- Applied Water
PROFESSIONAL PAPERS, AND CAPSTONES Science Vol. 7, pp. 2907–2918
[9] Gazzaz N.M., Yusoff M.K., Aris A.Z., Juahir H., Ramli M.F (2012).” [19] Thair Khayyun, Abdul Hameed M Jawad Al Obaidy, Ayad Mustafa
Artificial neural network modeling of the water quality index for (2014). “Prediction of Water Quality of Euphrates River by Using
Kinta River (Malaysia) using water quality variables as predictors”. – Artificial Neural Network Model (Spatial And T emporal Study)” --
Marine Pollution Bulletin Vol. 64, Issue. 11, pp. 2409-2420 International Research Journal of Natural Sciences Vol.2, No.3,
[10] M K Daud, Muhammad Nafees , Shafaqat Ali, Muhammad Rizwa, pp.25-38.
Raees Ahmad Bajwa, Muhammad Bilal Shakoor, Muhammad Umair [20] Vaishnavi V. Daigavane, Dr. M.A Gaikwad., “Water Quality
Arshad , Shahzad Ali Shahid Chatha , Farah Deeba ,Waheed Murad, Monitoring System Based on IOT(2017) ” -- Advances in Wireless
Ijaz Malook , Shui Jin Zhu (2017) .“Drinking Water Quality Status and Mobile Communications, ISSN 0973-6972..
and Contamination in Pakistan”. -- BioMed Res. Int., 2017, 7908183. [21] Yogalakshmi S. and Mahalakshmi A(2021). “Efficient Water Quality
[11] Jianhua Dong, Guoyin Wang, Huyong Yan, Ji Xu & Xuerui Zhang Prediction for Indian Rivers Using Machine Learning”-- Asian
(2015). “ A survey of smart water quality monitoring system” -- Journal of Applied Science and T echnolog, Vol. 5, Issue 1, pp 100 -
Environmental Science and Pollution Research volume Vol. 22, pp. 109
4893–4906 [22] Umair Ahmed , Rafia Mumtaz , , Hirra Anwar , Asad A. Shah ,
[12] Ali Najah Ahmeda, Faridah Binti Othman, Haitham Abdulmohsin Rabia Irfan and José García-Nieto (2019). “Efficient Water Quality
Afan, Rusul Khaleel Ibrahim, ChowMing Fai, Md Shabbir Hossain, Prediction Using Supervised Machine Learning”-- Environmental
Mohammad Ehteram, AhmedElshafie (2019). “Machine learning Chemistry of Water Quality Monitoring
methods for better water quality prediction” -- Journal of Hydrology
[13] Ishaani Priyadarshini, Ahmed Alkhayyat, Ahmed J.Obaid, Rohit
Sharma (2022). “Water pollution reduction for sustainable urban
development using machine learning techniques” – Cities Vol 130

978-1-6654-6408-6/23/$31.00 ©2023 IEEE 180

Authorized licensed use limited to: Ming Chi University of Technology. Downloaded on December 01,2024 at 13:27:12 UTC from IEEE Xplore. Restrictions apply.