[PPT] - Mode Detection, Age Pattern Transition, and its Consequences on PowerPoint Presentation

SLIDE 1

INDIAN INSTITUTE OF TECHNOLOGY BOMBAY, INDIA

Mode Detection, Age Pattern Transition, and its Consequences on Carbon Emissions

Team 5 Deepank Verma (D3) Neenu Thomas (D2) Omkar Deepak Karmarkar (D1)

SLIDE 2

Public modes for majority of work

trips

Private modes majorly for leisure

trips. .

Purpose

BACKGROUND: MODE CHOICE DETERMINANTS

39.17% & 25.47% of short trips

(less than 30 minutes travel) are made

n foot & by Car respectively.

Above 60% of long trips are made

n train.

Travel time

Lesser number of Private mode users during the working hours.

Walk and bicycle least preferred

for late night trips.

Trip start time Private mode for late night trips Walk and bicycle least preferred

for late night trips.

Trip end time Walk and bicycle more preferred

mode by female.

Public and private modes

more preferred mode by male.

Gender Most preferred mode

Children aged less than 15 years - Walk Working aged people – Public transport Elderly - Private mode

Age

SLIDE 3

PT Data
Total entries : 1461514
After cleaning : 1441493.
Data random split: 85:15, Training set

entries: 1225269, Test set entries: 216224

PP Data
Total entries : 1522
After cleaning : 1502
Data random split: 70:30, Training set

entries: 1051, Test set entries: 451

Target Variable (same)
Predictor Variables (same)

Data Analysis

Objective 1 : Comparing Mode Detection using Traditional and Machine Learning techniques.

Predictor Variables Purpose (5 categories) Gender (2 categories) Age (17 categories) Start Time End Time Travel Time Target Variable Mode (4 categories)

SLIDE 4

MULTINOMIAL LOGIT MODEL (MNL MODEL)

The most widely used mathematical form for choice

probabilities in behavioural travel demand analyses Key Strengths:

The MNL model is simple to perform
Computationally efficient
Permits a simple behavioural interpretation of its

parameters Key Weaknesses:

Independence of Irrelevant Alternatives (IIA) property
No correlation between error terms (i.i.d. errors)
Random taste variation can not be represented,

Log-Likelihood: -1464800 McFadden R2: 0.24345 Likelihood ratio: 942710 Testing Set: 216224 nos.

72885 63074 31675 48590 Predicted Actual Total samples

SLIDE 5

0.1 0.2 0.3 0.4 0.5 0.6 TravelTimeMin EndTimeHr StartTimeHr Age_10-14 Age_5-9 Purpose_Work Female Male Age_15-19 Purpose_Others Purpose_Ret. Home Purpose_Shopping Purpose_Leisure Age_20-24 Age_40-44 Age_75-79 Age_30-34 Age_60-64 Age_55-59 Age_45-49 Age_25-29 Age_80-84 Age_35-39 Age_65-69 Age_70-74 Age_85 more Age_50-54

RF

Random Forests (RF)

A decision tree is a decision support tool that uses a

tree-like graph or model of decisions.

RFs train each tree independently, using a random

sample of the data. Key Strengths:

RF is much easier to tune. Only two hyperparameters (a)

depth of trees (6) and (b) number of estimators (25).

More robust than a single decision tree, and less likely to
ver fit on the training data.
Does not require preparation of the input data.
Works with unscaled data and missing values.
Provides information on Feature importance.

Key Weaknesses:

For data including categorical variables with different

number of levels, RFs are biased against attributes with more levels.

72885 63074 31675 48590

Predicted Actual Total samples

Testing Set: 216224 nos.

SLIDE 6

0.05 0.1 0.15 0.2 0.25

TravelTimeMin EndTimeHr StartTimeHr Male Purpose_Work Purpose_Ret.… Purpose_Leisure Age_5-9 Purpose_Shopping Purpose_Others Age_10-14 Age_15-19 Age_70-74 Age_75-79 Age_20-24 Age_65-69 Age_80-84 Age_25-29 Age_35-39 Age_85 more Age_60-64 Age_40-44 Age_30-34 Age_45-49 Age_55-59 Age_50-54 Female

XGB

Extreme Gradient Boosting (XGB)

XGB build trees one at a time, where each new tree

helps to correct errors made by previously trained tree. Key Strength:

It performs the optimization which makes the use of

custom loss functions much easier.

Boosting focuses on unbalanced datasets by

strengthening the impact of the positive class.

Provides information on Feature importance.

Key Weaknesses:

Training generally takes longer because of the fact

that trees are built sequentially.

XGB is harder to tune than RF. Three parameters to

tune: (a) number of estimators (100), (b) depth of trees (6) and (c) Learning rate (0.1)

72885 63074 31675 48590

Actual Predicted Total samples

Testing Set: 216224 nos.

SLIDE 7

Artificial Neural Network (ANN)

ANN is based on a collection of connected units (nodes) called artificial

neurons, which loosely model the neurons in a biological brain. Key Strengths:

Ability to learn and model non-linear and complex relationships.
Efficiently processes large amount of training samples.

Key Weakness:

Hardware dependence
Unexplained behavior of the Model.
Comparatively sub-standard performance in smaller datasets.
Large number of hyperparameters to tune. (a) No. of Hidden Layers (3), (b)

Activation functions(Relu, Relu, Softmax), (c) Loss functions (Cross- entropy), (d) Dropout rate (0.8), (e) Optimizer (Adam), (f) Learning rate (0.01).

Testing Set: 216224 nos.

72885 63074 31675 48590

Predicted Actual Total samples

SLIDE 8

Evaluation Metrics

Overall Accuracy: The ratio of a total number of correctly identified

pixels to the total number of considered pixels.

The overall accuracy metrics is influenced by unbalanced and

prominent classes.

F1-score: is a harmonic mean of Precision and Recall.
Precision is the proportion of positive detections of the

classifier which were actually correct,

Recall refers to the proportion of actual positives which were

detected correctly.

Kappa index of Agreement is used in assessing the performance of

different models.

The Kappa value (k) of model classifier suggests that the classifier is

k*100 percent better than random assignment of classes.

F1-score Classes MNL RF XGB ANN Public 0.77 0.79 0.79 0.79 Private 0.46 0.5 0.52 0.51 Bicycle 0.03 0.02 0.3 0.34 Walk 0.52 0.51 0.58 0.59

Ov. Acc.

54.72 57.03 60.71 60.44 Kappa 0.368 0.388 0.451 0.45 F1-score Classes MNL RF XGB ANN Public 0.66 0.76 0.72 0.69 Private 0.53 0.63 0.66 0.55 Bicycle 0.46 0.52 0.57 0.52 Walk 0.49 0.49 0.51 0.48

Ov. Acc.

56.09 64.52 64.75 58.76 Kappa 0.379 0.491 0.499 0.406

PT data PP data

SLIDE 9

Potential Use Cases

Real-time congestion estimation/pricing : By the use

f mode detection models trained on telemetry data

from various smartphone applications. Advertising: Push Notifications influencing future mode choice behavior by cab aggregators based on trained model. Health: Evaluation of exposure to pollution by estimating the choice of mode.

SLIDE 10

Objective 2 : Analyzing the Mode Pattern considering the Shifting Age Composition

10

SLIDE 11

AGE PATTERN AND MODE CHOICE

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0-14 15-64 Above 65 Rail Bus Car Bicycle Walk

Mode use share

Higher share of on-road motorised modes by elderly people Increasing traffic volume of on-road vehicles Increasing carbon share by elderly Comfort and convenience of elderly people Measures to reduce carbon share by transportation Measures to ease the travel for elderly OBSERVATIONS

SLIDE 12

Hayashiya, H. Urban Rail Transit (2017) 3: 183. https://doi.org/10.1007/s40864-017-0070-4

MODE USE AND CHANGING EMISSION LEVEL

5.7 67.7 26.6

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

2010

Age-wise Share of Carbon Emission through Transport Modes

0-14 15-64 Above 65

Carbon Emission by Transport Modes Age wise share of different modes

Source: Hayashiya, H. Urban Rail Transit (2017) 3: 183. https://doi.org/10.1007/s40864-017-0070-4

0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Rail Bus Car Bicycle Walk 0-14 15-64 Above 65

SLIDE 13

CARBON EMISSION THROUGH TRANSPORT MODES

14.6 13.4 12.2 11.3 11 10.8 68.1 64.1 60 59.2 55.8 53.6 17.4 22.5 27.8 29.6 33.2 35.7

10 20 30 40 50 60 70 80 90 100 2000 2010 2020 2030 2040 2050

Changing Age Pattern

0-14 15-64 Above 65

Age Group

a. Changing Population trend
b. Same Mode pattern
c. Same Mode based emission per PKM

Scenario 1

a. Changing Population trend
b. Same Mode pattern
c. Changing Mode based emission per PKM

(Reference: Decadal average)

Scenario 2

a. Changing Population trend
b. Same Mode pattern
c. Changing emission per PKM in car while

emission per PKM in other mode are constant

Scenario 3

a. Changing Population trend
b. Shifting Car users to Public transport (Bus)
c. Changing Mode based Emission per PKM

Scenario 4

Source: National Institute of Population and Social Security Research

SLIDE 14

5 10 15 20 25 2010 2020 2030 2040 2050

CO2 Emission (g-CO2 /passenger-km)

Carbon Emission through Transport Modes

(in billions) Scenario 1 Scenario 2 Scenario 3 Scenario 4

CARBON EMISSION THROUGH TRANSPORT MODES

a. Changing Population trend
b. Same Mode pattern
c. Same Mode based emission per PKM

Scenario 1

a. Changing Population trend
b. Same Mode pattern
c. Changing Mode based emission per PKM

(Reference: Decadal average)

Scenario 2

a. Changing Population trend
b. Same Mode pattern
c. Changing emission per PKM in car while

emission per PKM in other mode are constant

Scenario 3

a. Changing Population trend
b. Shifting Car users to Public transport (Bus)
c. Changing Mode based Emission per PKM

Scenario 4

20 40 60 80 100 2010 2020 2030 2040 2050 2010 2020 2030 2040 2050 2010 2020 2030 2040 2050 2010 2020 2030 2040 2050

Percentage share Age Group

Age-wise Share of Carbon Emission through Transport Modes

Above 65 15-64 0-14 Scenario 1 Scenario 2 Scenario 3 Scenario 4

SLIDE 15

FLEXIBLE TRANSPORT SERVICES FOR EASE OF TRANSPORT FOR ELDERLY PEOPLE

Provide flexible transportation service (FTS) for elderly people in the city to access the metro station to improve their mobility. More detailed study of the FTS SYSTEMS

a b d e f

Route terminus Transfer point Scheduled bus stop Bus stop served by request only

c

Route deviation Fixed schedule and direction Point deviation No defined path or schedule, demand responsive and deviates routes between two points Demand responsive connectors Fixed route connectors Request stops Defined path and schedule Flexible route segments Fixed schedule and path Zone route One or more end points decided by passengers

a b c d e f

SLIDE 16