Taxi Demand Prediction using Machine Learning.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 05 | May 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 1527
Taxi Demand Prediction using Machine Learning.
P. Sudheer Benarji1, P. Sai Bharadwaj2, B. Neeha3, D. Srikanth4, V. Ankitha5
1Professor, VNR Vignana Jyothi Institute of Engineering & Technology, Hyderabad
2345Under Graduate Student, VNR Vignana Jyothi Institute of Engineering & Technology, Hyderabad
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Taxi demand prediction is the process of using
historical data to forecast future taxi requests in a particular
area. Managers may pre-allocate taxi resources in cities with
the aid of accurate and real-time demand forecasting, which
helps drivers find clients more quickly and cuts down on
passenger waiting times. This project is aimed to choose the
best model in predicting the taxi demand where we use
various Machine learning techniques such as regression
analysis and time series forecasting. Various baseline models,
including moving averages (simple, weighted, and
exponential), linearregressionwithgridsearch, randomforest
regressor with random search, and XGBoost regressor with
random search are used. We find out which model is more
suitable in predicting the output using the metrics we obtain.
Key Words: Linear RegressionwithGridSearchCV,Random
Forest Regressorwith RandomSearchCV, XGBoostRegressor
with RandomSearchCV.
1.INTRODUCTION
Taxi demand prediction is the process of using historical
data to forecast future taxi requests in a particular area.
Managers may pre-allocate taxi resources in cities with the
aid of accurate and real-time demand forecasting, which
helps drivers find clients more quickly and cuts down on
passenger waiting times.
In our project, we’ve used data on taxi rides in New York
city to to train and test the models using Linear regression,
Random Forest regressor and XGBoost regressor.
Along with the Linear regression and Random Forest
algorithms, we’ve also used the XGBoost algorithm.XGBoost
is a machine learning algorithm that is commonly used in
classification and regression problems. It is an ensemble
learning method that combines theweak predictionmodels ,
such as decision trees to create a stronger overall prediction
model. XGBoost has gained popularity due to its high
accuracy, scalability, and ability to handle missing data.
With our project, we get an understanding of which
model is best to predict the real time taxi demand and taxi
companies be able to tailor strategies to allocate resources
based on demand.
2. Literature Review
Multi-attention network-based graph prediction of
taxi demand: In order to better address the taxi demand
prediction problem, this study develops a Graph Multi-
Attention Network (GMAN), which tries to forecast the taxi
demands in every section of a roadnetwork.(Achieveda72%
Accuracy). Because only significant data needs to be learned
by the models, applying attention increases model accuracy
to extremely high levels. The Attention mechanism's
drawback is that it requires a lot of timeand is challenging to
parallelize
Taxi demand forecast usingtherandomforestmodel:
Decision trees are employed in the random forest. The term
"random" refers to our usage of a random bootstrap
sampling, and the term "forest" refers to the collection of
trees seen in decision trees. (Achieved 77% Accuracy).
excellent forecasting abilities that improve application
precision. Easy data preparation is made possible by not
requiring normalization.Generallyspeaking,thisalgorithmis
quick to train but takes a while to produce predictions after
training
Demand projection for taxis XGBoost algorithm-
based: The hot spot locations are identified and their
boundaries are drawn using the density-based DBSCAN
clustering technique, and thedemandforthehotspotareasis
predicted using the XGBoostalgorithm.XGBprovidesvarious
features, such as parallelization, cache optimization, and
more. Like any otherboostingmethod,XGBoostissensitiveto
outliers.
Taxi Demand Forecast Based on Regional
Heterogeneity Analysis and Multi-Level Deep Learning:
With the aid of the taxi zone clustering technique and
pairwise clustering theory, the Multi-Level RecurrentNeural
Networks (MLRNN) model is put out.(83.33% Accuracy
Attained). concentrates on how to exploit inter-zone
heterogeneity to enhance prediction. The use of MLRNN
results in high processing costs and greatercomplexitywhen
fitting data.
Probabilistic Taxi Demand Prediction with Bayesian
Deep Learning:ProposesaBayesiandeeplearningapproach
forprobabilistic taxi demand prediction.(AchievedAccuracy
of 83%). Estimates the uncertainty of predictions and
providesprobabilisticforecasts.Greatertechnicalcomplexity,
defining a prior distribution can be hard using Bayesian
statistics.
Prediction of Taxi Demand Using Ensemble Model:
Utilizes a point of interest (POI) to match taxi demand with a
location so that it can be studied using a different function.

This method is based on RNNs and XGBOOST. Achieved
Accuracy of 72%). It increases the accuracy, improved
resource allocation, effective dataanalysis.Limitedcoverage,
incomplete data, limited generalizability.
BRIGHT,Drift-Aware Demand Predictions for Taxi
Networks: Accurate forecasting of short-term taxi demand
amounts using a novel combination of time series analysis
techniques that can manage various sorts of concept
drift.(Achieved Accuracy of 78%). Could offer a range of
benefits, such as increased efficiency and revenue, and
improved customer service. The cost of implementing the
BRIGHT platform may be high, andongoingmaintenanceand
updates may also be required.
Taxi demand prediction using hybrid deep neural
networks: Hybrid deep neural network approach to predict
taxi demand based on a variety of factors. The authors
compare their approach to other machine learning
algorithms and find that their hybrid approach achieves the
highest prediction accuracy.(Achieved Accuracy of 80%).
Have shown to be effective at capturing complex patterns in
data, and a hybrid approach that combines different types of
networks can help improvepredictionaccuracy.Themodelis
too complex and begins to memorize the training data rather
than learning generalizable patterns.
Creating a Customised Transportation Model to
Predict Online Taxi Demand: a personalized demand
forecast model is suggested along with a broad demand
prediction for online cab hailing. It is suggested that a model
with two attentional blocks be used to account for both
temporal and spatial viewpoints.(Achieved Accuracy of
75.7%). By using user-specific data, personalized
transportation models can make more accurate predictions
for demand and supply of rides. Personalized transportation
models rely on user-specificdata, whichmaynotbeavailable
for all users. This can limit the effectiveness of the model and
make it less accurate.
Convolutional Spatiotemporal Multi-Graph Network
for Taxi Demand: They tested Deep TDP on two real-world
traffic datasets, and the results showed that it was effective
when compared to self- and other baseline
variations.(Achieved Accuracy of 80.5%). STMGCNs can
handle input data in various formats, such as graph-based
data and timeseries data, making them versatilefordifferent
types of applications. The model does not have the ability to
extract multi-scale correlations of non-adjacent frames.
A method for predicting taxi demand using an
ensemble: The authors discuss their ensemble technique,
which integrates many machine learning models, such as
random forest regression, linear regression, and support
vector regression.They made use of a variety of metrics,
including root mean squared error, mean squared error, and
mean absolute error. (Achieved Accuracy of 88%). Has
several potential benefits, including improved prediction
accuracy and robustness. It may also be more complex than
other approaches.
3. Existing system
There are several existing systems for taxi demand
prediction using machine learning. Here are a few examples:
Uber Movement: Uber Movement is a platform that uses
machine learning to predict the demand for Uber rides in
various cities. It provides historical and real-time data on
traffic patterns, events, and weather conditions to help
drivers optimize their routes and improve passenger wait
times.
NYC Taxi Demand Prediction: The New York City Taxi and
Limousine Commission (TLC) has developed a system for
predicting taxi demand in New York City using machine
learning algorithms. The system uses historical data on taxi
trips, weather, and events to forecast the number of taxi
requests in various parts of the city.
Didi Chuxing: Didi Chuxing is a Chinese ride-hailing
company that has developed a machine learning-based
system for predicting demand for its services. The system
uses real-timedata on trafficconditions, weather,andevents
to optimize ride allocation and reduce wait times for
passengers.
GrabTaxi: GrabTaxi is a Southeast Asian ride-hailing
company that uses machine learning to predict demand for
its services. The system uses historical data on riderequests,
traffic, and weather conditions to forecast demand and
allocate drivers accordingly
4. Proposed system
The purpose of this project is to build a model to analyze
data-patterns and predict the demand of taxis based on
number of requests in a given time period to help the taxi
companies to pre-allocate the resources optimally.
The objectives is to analyze taxi demand patterns, segment
requests based on pickup and drop-offpointsanddurationof
ride, extract features for machine learning models, and build
machine learning models using Linear regression, Random
Forest algorithm and XGBoost regressor to predict taxi
demand in the future

5. System Architecture
Fig 1: System Architecture
Client tier: This tier is responsible for presenting the user
interface of the system. It includeswebpages,mobileapps,or
other interfaces that allow users to interact with the system.
The client tier communicates with the logical tier to receive
and display data.
Logical tier: This tier contains the business logic and
application code of the system. It receives requests from the
client tier, processes them, and sends responses back. The
logical tier includes the data preprocessing, training models,
and prediction algorithms that predict taxi demand. This tier
also communicates with the data tier to access and retrieve
data.
Data tier: This tier is used for storing and retrieving data
from database. It includes the storage of data or data
warehouse that stores historical taxi demand data, weather
data, traffic data, and other relevant data sources used for
prediction.
6. Functionalities
i) Functionality of User Input:
The user input is taken through the keyboard, then it
analyses parameters and predicts the demand.
ii) Functionality of Pipeline Module:
Various machine learning algorithms are considered and
stored in the pipeline. All the machine learning algorithms
that are available in the pipeline are used to train the model
with the dataset.
iii) Functionality of Model Selection Module:
RMSE and R2 score values for the above models are
calculated. Then the model with the best figures ofRMSEand
R2 score is considered as the final model for the prediction.
iv) Functionality of Prediction:
Various data pointsare takenandtheaboveselectedmodelis
being used to predict the taxi demand .
7. UML Diagrams
Fig 3: Class Diagram
A class diagram is another name for a static diagram. It
shows the application's static view. A class diagram can be
used to generate an executable code directly from the
diagram for a software programmer as well as to visualize,
describe, and document various components of a system. A
class diagram describes the characteristics, actions, and
limitations of a class.
Fig 4: Sequence diagram
The lifelines are:
● User
● Prediction (ML) Model
● Data Set
A sequence diagram is a type of interaction diagram because
it shows how a group of actors communicate with one
another. Software engineers and business professionals use
these diagrams to understand the specifications for a new
system

Fig5: Use Case Diagram
A written example of how users will carry out tasks on your
website is known as a use case. It establishes how a system
reacts to a request from the perspective of a user. A series of
fundamental actions that begin with the user's aim and end
when that goal is accomplished define each use case.
8. RESULTS
Models are trained using the dataset and accuracy of each
model is analyzed to find out the best model for prediction
that can best tell the demand.
Fig – 6(a): Plot of Clusters
Fig – 6(b): Linear Regression Model Metrics
Fig – 6(c): RandomForest Regressor Metrics
Fig – 6(d)XGBoost Regressor Metrics
9. CONCLUSION
In conclusion, taxi demand prediction using machine
learning is a useful application that can help taxi companies
optimize their operations and improve customer
satisfaction. Use of machine learning provided many
advantages in predicting Taxi Demand. The model saved
time by preventing all the complex calculations and giving
the demand of taxi in particular area. We used several
essential attributes and regression techniques particularly
linear Regression, Random Forest Regressor, XGBoost
Regressor and K-Means for clustering of data points. We got
even more accuracy than other models.
REFERENCES
[1] Mohamed Hanafy, Assiut University” Predict Health
Insurance Cost by using Machine Learning and DNN
Regression”, Research gate, 348559741,2021.
[2] Shyamala Devi , Swathi Pillai , Vel Tech ”Linear and
Ensembling Regression Based Health Cost Insurance
Prediction Using Machine Learning “,Research gate,
353231212,2021.
[3] Ch Anwar Ul Hassan, Jawaid Iqbal“A Computational
Intelligence Approach for Predicting Medical Insurance
Cost Hindawi journal,1162553,2021.

[4] Kashish Bhatia, Shabeg Singh Gill, Navneet Kamboj,
Manish Kumar” Health Insurance Cost Prediction using
Machine Learning” IEEExplore,984201,2022.
[5] Chaparala Jyothsna, K. Srinivas, Bandi Bhargavi” Health
Insurance PremiumPredictionusing XGboostRegressor“
IEEExplore, 9793258,2022.
[6] Preet Jayendrakumar Modi, Vraj Jatin Naik “Insurance
Management with Premium Prediction “ IJRASET,2022
[7] Ghosh Madhumita “Health Insurance Premium
Predictionusing Blockchain Technology and Random
Forest Regression Algorithm” IJOEST article,346,2022.
[8] Keshav Kaushik , Akashdeep Bhardwaj ”Machine
Learning-Based Regression Framework to Predict
Health Insurance Premiums” NCBI,35805557,2022.
[9] Zhu, M., Li, Y., Li, L., & Li, Z. (2019). Time series
prediction of taxi demand based on deep learning. IEEE
Access, 7, 179647-179655.
[10] Yang, L., Zhang, B., Chen, Y., & Chen, L. (2018). A Deep
Learning Method for Taxi Demand Prediction. IEEE
Transactions on Intelligent Transportation Systems,
19(3), 782-791.
[11] Ma, T., Yang, Z., Wang, C., Hu, Y., Zhang, Y., & Liu, Y.
(2019). Deep Multi-View Spatial-Temporal Network for
Taxi Demand Prediction. IEEE Transactions on
Intelligent Transportation Systems, 20(5), 1745-1756.
[12] Yu, X., & Jiang, Y. (2019). A hybrid deep learning model
for taxi demand prediction. Transportation Research
Part C: Emerging Technologies, 101, 206-218.
[13] Zhang, S., Chen, J., & Chen, C. (2021). Hierarchical
Spatiotemporal Network for Taxi Demand Prediction.
IEEE Transactions on Intelligent Transportation
Systems, 22(5), 2918-2928.
[14] Zheng, X., Yang, X., & Sun, X. (2020). Deep Spatio-
Temporal Residual Networks for CitywideTaxiDemand
Prediction. IEEE Transactions on Intelligent
Transportation Systems, 21(3), 1032-1042.
[15] Li, Y., Li, Z., Li, S., Li, W., & Li, G. (2020). A novel urban
taxi demand prediction model using spatio-temporal
deep learning method. Journal of Cleaner Production,
259, 120877.
[16] Zhang, K., Li, K., Lin, P., & Liu, K. (2021). A newattention-
based neural network model fortaxidemand prediction.
TransportationResearchPartC:EmergingTechnologies,
126, 102988.

Taxi Demand Prediction using Machine Learning.

More Related Content

Similar to Taxi Demand Prediction using Machine Learning. (20)

More from IRJET Journal (20)

Recently uploaded (20)

Taxi Demand Prediction using Machine Learning.