Transportation Planning

Department of Civil Engineering

Maulana Azad National Institute of Technology Bhopal 1
 Traffic
Forecasting Models
Role of Models in Planning
 Forecasting, planning and goals
• Short-term forecasts

• Medium-term forecasts

• Long-term forecasts
 Forecasting data and methods
• Qualitative forecasting
• Delphi Method
• Forecasting by analogy
• Scenario forecasting
• Quantitative forecasting
• Time-series forecasting
• Predictor variables and time series forecasting

Demand=f(strength of economy, population, travel purpose, time of day, day of week,

error).
 Models
Commonly Used Forecasting Models as per present Research are:

1. Autoregressive Integrated Moving Average (ARIMA) Model for

Forecasting Future Traffic Growth Rate

2. Elasticity Model Relative to changes in traffic growth rate

3. Exponential Smoothening Model

4. Transport Demand Elasticity Model

 Basic steps in a forecasting

• Step 1: Problem definition

• Step 2: Gathering information

• Step 3: Preliminary (exploratory) analysis

• Step 4: Choosing and fitting models

• Step 5: Using and evaluating a forecasting model.

 ARIMA Models
❖ ARIMA is an acronym that stands for Auto Regressive Integrated Moving Average.

❖ Given a time series of data Xt , the ARMA model is a tool for understanding and predicting
future values in this series.

❖ This acronym is descriptive, capturing the key aspects of the model itself. Briefly, they are:

AR: Autoregression. A model that uses the dependent relationship between an observation
and some number of lagged observations.

I: Integrated. The use of differencing of raw observations (e.g. subtracting an observation

from an observation at the previous time step) in order to make the time series stationary.

MA: Moving Average. A model that uses the dependency between an observation and a
residual error from a moving average model applied to lagged observations.
ARIMA Models
• Role of Models

• Desirable features of a model

• Specification Calibration and Validation

• Fundamental Concepts (Utility maximization and

generalized cost; Equilibrium; Aggregation of individuals
decision to produce population estimates)
 Models available to the Transport Analysis
1. General Formula
2. Time series model
3. Averaging & Smoothing: forecasting without trend
4. Regression analysis
5. Matrix estimation Model
6. Elasticity Models
7. Simulation Models
General Formula
Time Series-based Forecasting
 A constant mean series
14.00

12.00

10.00

8.00
Series1
6.00

4.00

2.00

0.00
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

The above data points have been sampled from a normal

distribution with a mean value equal to 10.0 and a variance equal
to 4.0.
Time Series Models
Time series regression models
Averaging & Smoothing: forecasting without
trend
 Forecasting constant mean series:
The Moving Average model
The presumed model for the observed data:

D(t ) = D + e(t )
where
D is the constant mean of the series and
e(t ) is normally distributed with zero mean and some unknown
variance 
2

Then, under a Moving Average of Order N model, denoted as MA(N),

the estimate of D returned at period t, is equal to:
N −1
ˆ 1
D (t ) =
N
 D(t − i)
i =0
 The forecasting error

• The forecasting error

N −1
ˆ 1
 (t + 1) = D (t ) − D(t + 1) =
N
 D(t − i) − D(t + 1)
i =0

• Also

1 N −1 1
E[ (t + 1)] =  E[ D(t − i)] − E[ D(t + 1)] = ( ND ) − D = 0
N i =0 N

1 N −1 1 1 2
Var [ (t + 1)] = 2 Var [ D(t − i)] + Var [ D(t + 1)] = 2 N +  = (1 + )
2 2

N i =0 N N
 Forecasting error (cont.)
• (t+1) is normally distributed with the mean and variance computed
in the previous slide.

• Dˆ (t ) − D follows a normal distribution with zero mean and variance

2/N.
 Selecting an appropriate order N
• Smaller values of N provide more flexibility.
• Larger values of N provide more accuracy (c.f., the formula for the
variance of the forecasting error).
• Hence, the more stable (stationary) the process, the larger the N.
• In practice, N is selected through trial and error, such that it
minimizes one of the following criteria:
t
1
i) MAD(t ) = 
t − N i = N +1
|  (t ) |
ii) 1 t
MSD(t ) = 
t − N i = N +1
( (t ))2
iii)
1 t
 (t )
MAPE (t ) = 
t − N i = N +1 D(t )
 Demonstrating the impact of N on
the model performance
25.00

20.00

15.00
Series1
Series2
Series3
10.00

5.00

0.00
1

10

13

16

19

22

25

28

31

34

37

40
• blue series: the original data series.
• magenta series: the predictions of the MA(5) forecasting model.
• yellow series: the predictions of the MA(10) forecasting model.
• Remark: the MA(5) model adjusts faster to the experienced jump of the data mean value, but the mean estimates that it
provides under stationary operation are less accurate than those provided by the MA(10) model.
 Exponential Smoothing
With exponential smoothing the idea is that the most recent observations
will usually provide the best guide as to the future, so we want a weighting
scheme that has decreasing weights as the observations get older.
 Forecasting constant mean series:
The Simple Exponential Smoothing model
• The presumed demand model:

D(t ) = D + e(t )
where D is an unknown constant and e(t ) is normally distributed with zero mean
and an unknown variance  2 .
• The forecastDˆ (t ) , at the end of period t:

ˆ ˆ ˆ ˆ
D (t ) = aD(t ) + (1 − a) D (t − 1) = D (t − 1) + a[ D(t ) − D (t − 1)]
where (0,1) is known as the “smoothing constant”.
• Remark: The updating equation constitutes a correction of the previous
estimate in the direction suggested by the forecasting error, D(t ) − Dˆ (t − 1)
 Expanding the Model Recursion

ˆ (t ) = aD(t ) + (1 − a) D
D ˆ (t − 1)

ˆ (t − 2) =
= aD(t) + a(1− a)D(t −1) + (1− a)2 D
.................................................................................................
t −1
= a (1 − a) D(t − i) + (1 − a) Dˆ (0)
i t

i =0

 The impact of  and of Dˆ (0) on
the model performance
25.00

20.00

15.00 Series1
Series2
Series3
10.00 Series4

5.00

0.00
1

10

13

16

19

22

25

28

31

34

37

40
• Dark blue series: the original data series.
• magenta series: the predictions of the ES(0.2) model initialized at the value of 10.0.
• yellow series: the predictions of the ES(0.2) model initialized as 0.0.
• light blue series: the predictions of the ES(0.8) model initialized at 10.0.
• Remark: the ES(0.8) model adjusts faster to the jump of the series mean value, but the estimates that it provides under
stationary operation are less accurate than those provided by the ES(0.2) model. Also, notice that the effect of the initial
value is only transient.
 The inadequacy of SES and MA models for
data with linear trends
12

10

8
Dt
6 SES(0.5)
SES(1.0)
4

0
1 2 3 4 5 6 7 8 9 10

• blue series: a deterministic data series increasing linearly with a slope of 1.0.
• magenta series: the predictions obtained from the SES(0.5) model initialized at the exact value of 1.0.
• yellow series: the predictions obtained from the SES(1.0) model initialized at the exact value of 1.0.
• Remark: Both models under-estimate the actual values, with the most inert model SES(0.5) under-
estimating the most. This should be expected since both of these models (as well as any MA model)
essentially average the past observations. Therefore, neither the MA nor the SES model are appropriate for
forecasting a data series with a linear trend in it.
 Forecasting series with linear trend:
The Double Exponential Smoothing Model

The presumed data model:

D(t ) = I + T  t + e(t )
where

I is the model intercept, i.e., the unknown mean value for t=0,
T is the model trend, i.e., the mean increase per unit of time, and
e(t ) is normally distributed with zero mean and some unknown
variance  2
The Double Exponential Smoothing Model (cont.)

The model forecasts at period t for periods t+, =1,2,…, are

given by:
Dˆ (t +  ) = Iˆ(t ) + Tˆ (t ) 
with the quantitiesIˆ(t ) and Tˆ (t ) obtained through the following
recursions:
Iˆ(t ) = a  D(t ) + (1 − a)[ Iˆ(t − 1) + Tˆ (t − 1)]
Tˆ (t ) = b [ Iˆ(t ) − Iˆ(t − 1)] + (1 − b )  Tˆ (t − 1)
The parameters a and b take values in the interval (0,1) and are the
model smoothing constants, while the values Iˆ(0) and Tˆ (0) are the
initializing values.
 The Double Exponential Smoothing Model
(cont.)
• The smoothing constants are chosen by trial and error, using the
MAD, MSD and/or MAPE indices.
• For t→, Iˆ(t ) → I and Tˆ (t ) → T
• The variance of the forecasting error,   , can be estimated as a
2

function of the noise variance 2 through techniques similar to

those used in the case of the Simple Exp. Smoothing model, but in
practice, it is frequently approximated by
ˆ 2 = 1.25MAD(t )
where
MAD(t ) = g  (t ) + (1 − g ) MAD(t − 1)
for some appropriately selected smoothing constant g(0,1) or by
ˆ 2 = MSD(t )
 DES Example
12

10

8
Dt
6 DES(T0=1)
DES(T0=0)
4

0
1 2 3 4 5 6 7 8 9 10

• blue series: a deterministic data series increasing linearly with a slope of 1.0.
• magenta series: the predictions obtained from the DES(0.5;0.2) model initialized at the exact value of 1.0.
• yellow series: the predictions obtained from the DES(0.5;0.2) model initialized at the value of 0.0.
• Remark: In the absence of variability in the original data, the first model is completely accurate (the blue and
the magenta series overlap completely), while the second model overcomes the deficiency of the wrong initial
estimate and eventually converges to the correct values.
 Time Series-based Forecasting:
Accommodating seasonal behavior
The data demonstrate a periodic behavior (and maybe some
additional linear trend).

Example: Consider the following data, describing a quarterly

demand over the last 3 years, in 1000’s:

` Year 1 Year 2 Year 3

Spring 90 115 120
Summer 180 230 290
Fall 70 85 105
Winter 60 70 100
Total 400 500 615
 Seasonal Indices
Plotting the demand data:

350

300

250

200
Series1
150

100

50

0
0 2 4 6 8 10 12 14

Remarks:
• At each cycle, the demand of a particular season is a fairly stable percentage of the total demand over the cycle.
• Hence, the ratio of a seasonal demand to the average seasonal demand of the corresponding cycle will be fairly
constant.
• This ratio is characterized as the corresponding seasonal index.
 A forecasting methodology
Forecasts for the seasonal demand for subsequent years can be obtained by:
i. estimating the seasonal indices corresponding to the various seasons in the
cycle;
ii. estimating the average seasonal demand for the considered cycle (using, for
instance, a forecasting model for a series with constant mean or linear trend,
depending on the situation);
iii. adjusting the average seasonal demand by multiplying it with the
corresponding seasonal index.

Example (cont.):

Year 1 Year 2 Year 3 SI(1) SI(2) SI(3) SI

Spring 90 115 120 0.9 0.92 0.78 0.87
Summer 180 230 290 1.8 1.84 1.88 1.84
Fall 70 85 105 0.7 0.68 0.68 0.69
Winter 60 70 100 0.6 0.56 0.65 0.6
Total 400 500 615 4 4 4 4
Average 100 125 153.75
1. Estimate 5 years moving average and
10 years moving average.
2. Estimate exponentially smoothened
with α =0.3 and α = 0.5
3. Estimated the Flow value for next 5
years.
 Year 1935 1936 1937 1938 1939

Car
1648 1665 1627 1791 1797
ownership

Calculate forecasts for the next three years (1940–1942)

Year 1964 1965 1966 1967 1968

Traffic 467 512 534 552 545

Calculate forecasts for the next three years (1969–1971)

 Multiple Linear Regression
• The basic model:
D = b0 + b1  X1 + ... + bk X k + e
where
• Xi, i=1,…,k, are the model independent variables (otherwise known as the explanatory
variables);
• bi, i=0,…,k, are unknown model parameters;
• e is the a random variable following a normal distribution with zero mean and some
unknown variance 2.

• D follows a normal distribution N ( D ,  2 ) where

D = b0 + b1  X 1 + ... + bk  X k
• We need to estimate <b0,b1,…,bk> and 2 from a set of n observations

{ D j ; X 1 j , X 2 j ,..., X kj , j = 1,..., n}
 Estimating the parameters bi
• The observed data satisfy the following equation:

 D1   1 X 11 ... X k1  b0   e1 
D   1 X ... X k 2   b1  e2 
 2 =  12
+
 ...  ... ... ... ...   ...   ... 
      
 Dn   1 X 1n ... X kn  bk  en 
or in a more concise form
d = X b + e
• The vector
e = d − X b
denotes the difference between the actual observations and the corresponding mean
values, and therefore, b̂ is selected such that it minimizes the Euclidean norm of the
resulting vector . eˆ = d − X  bˆ
• The minimizing value for b̂ is equal to bˆ = ( X T X ) −1 X T d
• The necessary and sufficient condition for the existence of( X T X ) −1 is that the columns of
matrix X are linearly independent.
 Characterizing the model variance
• An unbiased estimate of 2 is given by

SSE
MSE = (Mean Squared Error)
n − k −1
where

SSE = eˆT  eˆ = (d − X  bˆ)T (d − X  bˆ) (Sum of Squared Errors)

goodness of fit • The quantity SSE/2 follows a Chi-square distribution with n-k-1 degrees of freedom.

• Given a point x0T=(1,x10,…,xk0), an unbiased estimator of D ( x0 ) is given by

Dˆ ( x0 ) = bˆ0 + bˆ1  x10 + ... + bˆk  xk 0

• This estimator is normally distributed with mean D ( x0 ) and variance  2 x0T ( X T X ) −1 x0

• The random variable Dˆ ( x0 ) can function also as an estimator for any single observation
D(x0). The resulting errorDˆ ( x0 ) − D ( x0 )will have zero mean and variance  2 [1 + x0T ( X T X ) −1 x0 ]
 Assessing the goodness of fit
• A rigorous characterization of the quality of the resulting approximation can be obtained
through Analysis of Variance, that can be traced in any introductory book on statistics.

•A more empirical test considers the coefficient of multiple determination

SSR
R =2

SYY
where n
SSR = bˆT ( X T d ) − nd 2 =  ( Dˆ j − d ) 2
n j =1
1
d =  Dj
and
n j =1

SYY = SSE + SSR

• Remark: A natural way to interpret R2 is as the fraction of the variability in the observed
data interpreted by the model over the total variability in this data.
 Multiple Linear Regression and
Time Series-based forecasting
• The model needs to be linear with respect to the parameters bi but not the explanatory
variables Xi. Hence, the factor multiplying the parameter bi can be any function fi of the
underlying explanatory variables.

• When the only explanatory variable is just the time variable t, the resulting multiple linear
regression model essentially supports time-series analysis.

• The above approach for time-series analysis enables the study of more complex
dependencies on time than those addressed by the moving average and exponential
smoothing models.

• The integration of a new observation in multiple linear regression models is much more
weighty than the updating performed by the moving average and exponential smoothing
models (although there is an incremental linear regression model that alleviates this
problem).
 Confidence Intervals
• Given a random variable X and p(0,1), a p100% confidence
interval (CI) for it is an interval [a,b] such that
P ( a  X  b) = p
• Confidence intervals are used in:
i. monitoring the performance of the applied forecasting model;
ii. adjusting an obtained forecast in order to achieve a certain
performance level

• The necessary confidence intervals are obtained by exploiting

the statistics for the forecasting error, derived in the previous
slides.
Exponential smoothing is a rule of thumb technique for smoothing time
series data using the exponential window function.

Whereas in the simple moving average the past observations are weighted
equally, exponential functions are used to assign exponentially decreasing
weights over time.

It is an easily learned and easily applied procedure for making some

determination based on prior assumptions by the user, such as seasonality.

Exponential smoothing is often used for analysis of time-series data.

 Elasticity Model
• As per this model, the elasticity of demand with respect to a certain
variable, such as travel cost, is defined as the rate of change of
demand with respect to that variable, normalized by the current
levels of demand and the variable in question.

• The elasticity model considers demographic and cost factors, show

the impact on land use and travel frequency of improving the traffic
conditions; discuss the induced traffic with time effect
two stations
 Elasticity Model

Ln P = -11.304 + 1.8976 Ln GDP

Here, P is the traffic volume in PCU/day and GDP is the gross domestic

product or constant price ( in Rs. Crores) of the region.

 Exponential Smoothening Method
• With exponential smoothing the idea is that the most recent observations will
usually provide the best guide as to the future, so we want a weighting scheme
that has decreasing weights as the observations get older.

• Simple exponential smoothing model is only good for non-seasonal patterns

with approximately zero-trend and for short-term forecasting because if we
extend past the next period, the forecasted value for that period has to be
used as a surrogate for the actual demand for any forecast past the next period
Matrix Estimation Models
Simulation Models
Travel Demand Forecasting