IRJET- Prediction of Cab Demand using Machine Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 03 | Mar 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 6961
Prediction of cab demand using machine learning
Abdul Rahuman Aslam M A1, Gobinathan V2, Krishnan K3, Rajasekaran G4
1Student, Easwari Engineering College, Bhrathi Salai, Ramapuram, Chennai – 600089.
2 Student, Easwari Engineering College, Bhrathi Salai, Ramapuram, Chennai – 600089.
3
Student, Easwari Engineering College, Bhrathi Salai, Ramapuram, Chennai – 600089.
4Assistant Professor, Easwari Engineering College, Bhrathi Salai, Ramapuram, Chennai – 600089.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The cab service industryisboomingforthe
last couple of years and it is expected to grow in the
near future. Taxi drivers need to decide where to wait
for passengers as they can pick up someone as soon as
possible. Passengers also prefer a quick taxi service
whenever needed. The control centre of the taxi service
decides the busy area to be concentrated. In the existing
system, sometimes the taxis were scattered across the
larger area missing the time based busy area like
Airport, Business area, school area, Train stations etc.
Effective taxi allocation can help both drivers and
passengers to minimize the wait-timetofindeachother.
In the proposed system, the future demand can be
predicted using Recurrent Neural Networks based
model that can be trained with given historical data. It
can serve more customers in a short time by organizing
the availability of taxi. The data set includes GPS
location and other properties of the taxi like drop point,
pickup point etc. This model is used to predict the
demand for aparticular time in different areas of the
city.
Key Words: Taxi demand prediction, Recurrent neural
network.
1.INTRODUCTION
Taxi drivers need to decide where to wait for the
passengers so that they can pick someone quickly.
Similarly, passengers also need to find their cabs
quickly. Dispatching the taxi efficiently help both the
customers and drivers. Effective dispatching of taxi
helps to reduce waiting time for customers, as well
as drivers.
A driver will not have enough information about
where to wait in order to get passengers quickly. A
taxi centre can organize and send the required
number of taxis to the area based on the historical
data. The historical data uses Global Positioning
System (GPS) and predict future demand. In Tokyo,
today this system is reducing the waiting time for the
customers, quickly respond to the sudden change of
demands and it bridges the gap between the
experienceddriversandnovicedrivers.Thesebenefits
allow the cab service to achieve maximum benefit.
A real-time taxi demand prediction is proposed here
andinthissystem,historicaldataisusedtopredictthe
future demand for taxis in a particular place at a
particular time. Some of the real-time objectives
include managing fleet of taxi to crowded area,
effective utilization of resources to reduce waiting
time, server more customers in a short time by
organizingavailabletaxi.OursystemusesGPSlocation
andother properties ofthetaxilikepickuppoint,drop
point etc. to predict taxi demand. A model is trained
using a recurrent neural network. The recurrent
neural networks are used in speech recognition
software. Therecurrentneuralnetworksareusedfor
sequential data. It is being used in Google’s voice
search and Apple Siri personal assistants. This
algorithm is the achievement of deep learning in past
years.Recurrentneuralnetworksproducepredictions
result for the sequential data.
Taxi demand prediction is a time series analysis
problem. Itisafeedforwardworkneuralnetworkand
the information moves from one network to another
network and from the input layer to the output layer
through the hidden layer. The difference between the
normal neural network and the recurrent neural
network is that in the recurrent neural network, the
information cycles through the loop.
In the system, the recurrent neural network is used
andPythonlanguage ispreferredbecauseithasavast
collection ofmachine learning libraries. The data set
might contain empty values, negative values or error.
Data set is cleaned in the preprocessing. The
preprocessing methods involve removing records
which are not complete. Once the cleaned data set is
available it is prepared to be fed to the machine

learning algorithm. Recurrent Neural Networks take
the previous node output or hidden states as inputs.
RNNs are useful as their intermediate values (state)
can storeinformationaboutpreviousinputsforatime
interval. The main feature of a Recurrent Neural
Network (RNN) is that the network possess at least
one feedback connection, so the activations can flow
round in a loop-wise manner.
That enables the networks to learn sequences and to
do temporal processing, e.g., perform sequence
recognition/reproduction or temporal
association/prediction. Recurrent neural network
architectures can have many variated forms. One
common
The type that consists of a standard Multi-Layer
Perceptron (MLP) plus added curves. These can
escapade the powerful non-linear mapping
capabilities of the MLP, and also have some form of
memory. Since one can think about recurrent
networks in terms of their properties as dynamical
systems, it is needed to ask about their stability,
observability and controllability:
Stabilityenterprisestheboundednessovertimeofthe
network outputs and the response of the network
output to minute changes to the weights or network
inputs. Controllability is concerned with whether it is
possible to control the dynamic behaviour patterns.
An RNN is said to be controllable if an original state is
steerable to a desired state within a finite number of
sequential time steps.Observabilityisconcernedwith
whether it is possible to observe the results of the
control that is applied. A recurrent network is
observable if the state of the network can be dogged
from a fixed set of input and output measurements.
The network input is the current taxi demand, while
the output is the demand in the next time-step. The
reason recurrent neural network is used is that it can
be trained to store all the relevant information in a
sequence to predict particularoutcomes in the future.
Itisatimeseriesforecastingproblemtopredictfuture
demand. Hence, a sequential algorithm is used. It is
desired to predict taxi demand in small areas so that
the drivers know exactly where to go. The system is
trained with the data set and create the model for
future prediction.
A graph isplottedforthefuturepredictionforthenext
time slot and the area to be crowded. This machine
learning model predicts the future demand area in a
city based on neural network and the drivers were
taken to wait in the area where the system identified
as demand area.
2. RELATED WORKS
Fei Miao et al [1] proposed a modern robust
transportationsystem that senses datacollectedfrom
transportation systems that help in analyzing the
passenger demands. Prediction Methods on taxi-
passenger demand were travel time and travelling
speed according to traffic monitoring data have been
developed. In their proposed model Robotic mobility-
on-demand systems that minimize the number of
rebalancing trips and best parking systems that
allocate resource based on a driver’s payments.
These kinds of algorithms aim to reduce long mile or
to minimize customers’ waiting time have been
developed. Although optimizationinrobust approach
aims to minimize the worst case cost under all
possible random parameters, it results in average
system performances. For a taxi dispatch system, it is
essential to address the compensation between the
worst case and the average dispatch costs under
uncertain demand. Estimations show that under the
robust dispatch framework we design, the average
demand-supply ratio imbalance is reduced by 31.7%,
and the average total idle driving distance is reduced
by 10.130% or about 20 million miles in total in one
year.
Mohammad Saiedur Rahaman et al [2] defined the
neighbourhoodidentificationprobleminthepresence
of a large number of heterogeneous contextual
modules. It codifies research as a problem of less wait
time prediction for taxi drivers at airports and
investigate heterogeneous elements related to time,
weather, flight arrivals and taxi trips. Taxis are
regarded as the easiest mode of transport for
transfer between the airport and the city. The queue
managers continuously monitor the concurrent
queues related to taxis and passengers and instruct
taxi drivers to join the passengers at the terminal
when there is higher demand. Toensure the seamless
operation of this process, the queue manager
estimates the demand for taxis in future. Airport
satisfaction ratings depend on the proper
management of both passenger queues and taxi.
Aimingtomaintaindemand-supplysymmetryoftaxis,

the airport transport managers employ an approach
where it requires extended humanintervention. They
examine the corresponding p-values, which show the
statistical significance of this difference. In this D-
value is around 0:45 and p-value are less than 0.001.
This means that neighbourhoods are different
statisticallyfor differentk-values. Ifmedianandmean
denote the improvement shown by the DIbiased
weights method over the baseline method in terms of
medianandmeanofpredictionerrorsrespectivelyfor
different k-values, the correlation between the
corresponding D-values and the prediction errors,
d_error(Median)andd_error(Mean)canbemeasured
to show the relationship between the improvement
inthedensequalityneighbourhoodandtheprediction
accuracy is improved. This paper shows Pearson’s
correlation scores of 0.393 (with mean) and 0.484
(with median). They argue that the quality of the
neighbourhood they identified is significantly
improved by the consideration of relevant
heterogeneous contextual factors, thus the
performance isboosted (i.e. mean prediction error is
less than 0.09 and the median prediction error is
less than 0.06).
DeshengZhang et al [3]states that theexistingsystem
of data collection is offline and collected by manual
investigations and it may result in inaccurate data for
real-time analysis. To address this, they have used a
model called Dmodel, employing roving taxicabs and
using them as real-time mobile sensors. By
implementing this, they can infer arriving passenger
moments by investigating the logical information.
They used 450GB dataset of 14,000 taxicabs for a half
year and it achieves 83% accuracy and outperforms
the statistical model by 42%. Passenger demand
prediction may be halted by bad weather, special
events or accidents. They provide two parts of the
solution for this. First, they mine a large dataset of
historical data, consisting of taxi passengers and their
related pickups. Second, to address the real-time
dynamics, they consider thousands of taxi cabs and
use them as real-time mobile sensors. It is possible
because they use GPS data in dense urban areas. The
front-end systemandtheback-enddispatchingcentre
form a network called roving sensor network. The
used Dmodel observes hidden contexts to infer
demand based on historical and real-time taxi cab
data. Dmodel analyses both the offlineanalysisofdata
and real-time data collected from roving taxis or
sensor network formed by them. They used a novel
parameter,namelythepickuppattern.Dmodelutilizes
the real-time pickup pattern so that the model can
select customized and compact training data to
increase the accuracyof inference. The use ofDmodel
yields 83% accuracy and outperforms statistical
model by 42%. They used a Hidden Markov Chain for
implementation. The Dmodel- based dispatching
outperforms basic and SDD based by 11% on an
average. This is due to the accurate inference by
Dmodel.
LuisDamasetal[4]proposedanovelmethodologyfor
predicting the spatial distribution of taxi–passengers
for a short time horizon using streaming input data.
First, the information was accumulated into a
histogram time-series. Then, three time-series
forecasting techniques were combined to emerge a
prediction. Experimental tests were conducted using
the online data set that is transmitted by 441 vehicles
of a fleet running in the city of Porto, Portugal. The
resultselaboratedsothattheframeworkproposedcan
provide effective learning into the spatio-temporal
distributionoftaxi–passengerdemandforahorizonof
30minutes.Thispaperfocusesonthereal-timechoice
problem of which is the best cab service stand to go to
after a passenger drop-off (i.e., the stand where
another passenger can be picked up within a short
span of time). An intelligent approach regarding this
flaw will improve network reliability for both
companies and customers; an intelligent distribution
of vehicles throughout stands will minimize the
average waitingtime to pickup a passenger, whilethe
distance travelled will be more profitable.
Furthermore, whenever they need a taxi, passengers
will also experience a lesser waiting time to get a
vacanttaxi.Themajorcontributionofthispaperfacing
this is to build predictions on the spatio-temporal
distribution of the taxi–passenger demand using
only streaming data. As a result, the model that is
presented has beenable to predict thetaxi–passenger
demand at each one of the 64 taxi stands for 30
minutes period intervals. The model has presented a
more than satisfactory performance, correctly
predicting the 506 874 tested services with an
aggregated error measurement lower than 26.11%.
Biao Leng et al [5] elaborated the battle between two
taxicompaniesinChinanamelyDidiandKauidadithat
occurred in 2014. The two companies are backed
up by internet giants like Tencent and Alipay. These
companies promoted the taxi drivers by giving them

incentives for each ride and also allowed the users
also to use their application by giving frequent
discounts and offers and also promoted payment
throughthemobilephone.Inthispaper,theycollected
a 37-day trip data and use 9000 entries in Beijing.
For the first 18 days there was no battle and for the
next 19 days, the battle was competitive. The spatial-
temporal data are studied and based on the
comprehensive analysis, benefits and drawbacks are
discussed. Taxi drivers welcomed this battle and
startedtoaccepteverypassengertheycansothatthey
can achieve maximum benefits from incentives.
Customers started to complain that they cannot find
any taxis without using these apps. So, offline taxi
services may get affected. So, this paper addresses
that offline cab services may get affected. These
companies wanted to increase the usage of mobile
payments and hence used money promotion.
Indirectly, money promotion increased cab service
indirectly. Our paper helps both online and offlinecab
services, as it is purely dependent on historical data
based on the Global Positioning System.
Nicholas Jing Yuan et al [6] proposed a recommender
system for both cab drivers and people expecting to
take a cab, using the knowledge of 1) passengers’
mobility patterns and 2) taxi drivers’ picking-
up/dropping-off behavioural patterns learned from
the GPS trajectories of taxicabs. In their method, they
learn the above-mentioned knowledge (probability
representation) from GPS trajectories of taxis. Then,
they feed knowledge into a probabilistic model that
calculate the profit of the candidate locations for a
particular cab driver based on where and when the
driverrequeststherecommendation.Theyproposeda
defined approach to detect parking places based on a
large number of GPS trajectories generated by taxis,
where the parking places stand for the locations
where cab drivers usually wait for passengers with
theircars parked. Theyinventedaprobabilistic model
to formulate the time-dependent taxi behaviors
(picking-up/dropping-off/cruising/parking) and
enable a city recommendation system for both
passengers and taxi drivers. They also improve the
taxi recommender by considering the time-varying
queue length at the parking places. Forbothofthetaxi
recommender and the passenger recommender, they
constructed a model consolidating day of the week
and historical weather conditions to the varying pick-
up/drop-off behaviors. They also perform large-scale
evaluations including in-the-field user studies to
validate their system. They also built the
recommender system with a data set generated by
12,000 taxicabs in a period of 110 days and evaluated
the system by large-scale experiments including a
series of in-the-field studies. As a result, the taxi
recommender predicts accurately the time-varying
queue in parking lots and the recommender
successfully suggests the segment of the road where
users can easily find vacant taxis. In the future, they
alsoplantodeployourrecommenderintherealworld
so as to further validate andincreasethe effectiveness
and robustness of thissystem.
Jun Xu et al [7] proposed a system in which they used
recurrent neural networks to predict the cab demand
for the future based on the historical data which has
GPS. They used Long Short Term Memory (LSTM) to
predict the future demand. They used the LSTM in
order to store the sequential data and predict the cab
demand for future. LSTMs are used in handwriting
recognition, Natural language Processing etc. LSTM
used in this model uses some gating mechanism to
store the previous value. In thismodel,theyhaveused
a Mixed Density Network (MDN) along with Long
Short Term Memory (LSTM). This paper can achieve
83% accuracy in predicting the future cab demand
and it contains 17% prediction error.
Tian he et al [8] proposed a receding horizon control
(RHC) framework to dispatch cabs, which
incorporates highlyspatiotemporallycorrelatedreal-
time Global Positioning System and demand/supply,
models(GPS)locationandoccupancyinformation.The
target includes matching spatiotemporal ratio
between demand and supply for service quality with
less usage of current and assumed future taxi idle
drivingdistance.TheydesignedanRHCframeworkfor
large-scale taxi dispatching. They look at both current
and future demand, saving costs under constraints by
involving expected future idle driving distance for
rebalancingsupply.Theframeworkcoverslarge-scale
data in real-time control. Sensing this kind of data is
used to build predictive passenger demand, taxi
mobility models, and serve as real-time feedback for
RHC. Their Future plan is to develop a privacy-
preservingcontrolframeworkwhendataofsomecabs
are not shared with the dispatch centre. Extensive
trace-driven analysis with a data set containing taxi
operational records in San Francisco, CA, USA, shows
that our end solution reduces the average total idle
distance by 52.01%, and reduces the supply-demand

ratio error across the city during one experimental
time slot by 45.02%.
3. SYSTEM ARCHITECTURE
The real wordtaxi-tripdatasetiscollected.The
collected data is pre-processed and the data
preprocessing is a data mining technique that involves
transforming raw data into an understandable format.
Real-world data is often incomplete, inconsistent,
and/or lacking in certain behaviors or trends, and is
likely to contain many errors. The major tasks of the
data preprocessing involves data cleaning, data
integration, data transformation, data reduction etc.
The missingvaluesareidentifiedanditisreplacedwith
mean values.Thecleaneddatasetisfedtotherecurrent
neural network model and is it trained using historical
data which consists of date, time, pickup location,
weather etc. and the future demand for the taxi is
predicted. Recurrent Neural Networks (RNN) are a
powerful and robust type of neural networks and
belong to the most promising algorithms out there at
the moment because they are the only ones with an
internal memory. The predicted demand for the cab is
visualized using a graph.
Fig. 3.1 System Architecture
4. SYSTEM IMPLEMENTATION
The dataset consists of three columns and
30,426 entries. The minimum index value and
maximum index values are found. The maximum and
minimum indices are used to format the dataset. The
output consists of a graph consisting of missing values
in the dataset corresponding to the given data with
time in a one-hour interval in x-axis and number of
pickups in the y-axis. It also consists of a graph
consisting of bookings with time in a one-hourinterval
in x-axis and number of pickups in the y-axis. Once,
missing values are found, the mean value for the
columns is found and this value is replaced in the place
of missing entries. There are a lot of options for
replacing missing values but the simplest andthemost
common one is replacing them with the mean of all the
entries under the respective column. At the end of the
data preprocessing, a cleaned data set with no missing
values is found and all the missing values in the data
set were replaced by the mean of the respective
columns. The cleaned dataset is important because
there may be some irrelevant data in the dataset that
may cause prediction error and causes inaccurate
results. The Comma Separated file (CSV) is collected
from the usage of the user and perform cleaning and
processing, as a result of a cleaned data set is achieved.
The cleaned data set is used for the next stage in
predicting the taxi demand the recurrent neural
network is applied to predict the future cab demand.
.
Fig. 4.1 Bookings graph
The figure 4.1 shows the plotted graph using
matplotlib. pyplot, where it is plotted as a ratio of time
in hours and number of pickups. The graph shows the
number of pickups, at a particular date.
The Vanilla Neural Networks or Conventional
Neural Networks accept only a fixed size inputs and
produce fixed sizeoutputs.Itcannotworkforsequence

of inputs and sequence of outputs. Recurrent Neural
Network is a class of Artificial Neural Network (ANN).
The network input is the current taxi demand and
other relevant information while the output is the
demand in the next time-step. The reason a recurrent
neural network is used is that it can be trained to store
all the relevant information in a sequence to predict
particular outcomes in the future. In addition, taxi
demand predictionisatimeseriesforecastingproblem
in which an intelligent sequence analysis model is
required. It is desired to predict taxi demand in small
areas so that the drivers know exactlywheretogo.The
system is trained with the dataset and the model is
used for future demand prediction.
Fig. 4.2 Artificial Neural Networks
Recurrent Neural Network considers previous input
and the current inputtocomputethefutureoutput.For
taxi demand, we use the RNN model because taxi data
is a sequential data and it is a time series forecasting
problem. Unlike, traditional neural network RNN has
memory for storing relevant parts of the input and use
it for prediction of the future demand. Recurrent
Neural Networks take the previous node output or
hidden states as inputs. RNNs are useful as their
intermediate values (state) can store information
about previous inputs for a time interval. The main
feature of a Recurrent Neural Network (RNN) is that
the network possesses at least one feedback
connection, so the activations can flow roundinaloop-
wise manner. That enables the networks to learn
sequences and to do temporal processing,e.g.,perform
sequence recognition/reproduction or temporal
association/prediction. Recurrent neural network
architectures can have many variated forms. One
common type that consists of a standard Multi-Layer
Perceptron (MLP) plus added curves. These can
escapade the powerfulnon-linearmappingcapabilities
of the MLP, and also have some form of memory. Since
one can think about recurrent networks in terms of
their properties as dynamical systems, it is needed to
ask about their stability, observability and
controllability. Stability enterprises the boundedness
over time of the network outputs and the response of
the network output to minute changes to the weights
or network inputs. Controllability is concerned with
whether it is possibletocontrolthedynamicbehaviour
patterns. An RNN is said to becontrollableifanoriginal
state is steerable to a desired state within a finite
number of sequential time steps. Observability is
concerned with whether it is possible to observe the
results of the control that is applied. A recurrent
network is observable if thestateofthenetworkcanbe
dogged from a fixed set of i/p and o/pmeasurements.
For predicting the future taxi demand, a dataset which
consists of date and time, number of pickups, number
of passengers, maximum temperature, minimum
temperature, humidity and ,wind speedisused.RNNis
called recurrent because it performs the same
computation for every input element and each output
is conditioned on previous input element.
Fig. 4.3 Recurrent Neural Network
Long short-term memory is RNNarchitecturemodel.It
can be used for handwriting recognition, speech
recognition etc. LSTM consists of an input gate, output
gate and a forget gate. LSTMs are explicitly designed in
order to avoid the long term dependency problem.
Traditional recurrent neural network has a chain of
repeating modules. But, in LSTM this structure is
different.

Fig. 4.4 Repeating Modules in standard RNN
Fig. 4.5 Repeating Modules in LSTM
LSTMs also have this chain like structure, but the
repeating module has a different structure. Instead of
having a single neural network layer, there are four,
interacting in a very special way.
Fig. 4.6 LSTM Model loss
The figure 4.6 shows the model shows the Long Short-
Term Memory (LSTM) model loss graph, whose
accuracy can be increased in each iteration. In the
neural network, when each data passes from the input
layer to the output layer, through the hidden layer the
accuracy of the model increases and each timethedata
passes the accuracy increases.
Figure 4.7 shows the number of days with greaterthan
800 pickups, which indicate peak hours. Peak hours
indicate the time in which the demand for the taxi is
high.
Fig. 4.7 Predicted Number of taxi pickups vs. Actual
Number of taxi pickups
Fig. 4.8 Predicted Number of taxi pickups vs. Actual
Number of taxi pickups

Figure 4.8 shows the graph plotted between predicted
number of taxi pickups and actual number of taxi pickups
from the historical data.
Fig 4.9 Estimated Number of Taxi Pickups
Fig 4.9 shows the estimated number of taxi pickups
for the next six hours.
5. CONCLUSION
The proposed system is a sequential learning model
with recurrent neural network for predicting the taxi
demandin different areas inthecity.Learningfromthe
past historical data, the demand prediction is done for
the location. Three Years data of the Airport is used to
train our model. This model gives the predictionoftaxi
demand for hourly basis and a particular time.
This work can be extended in the future by adding
more input such as holidays, festivals etc. Taxis can be
organized and send based on the prediction of the
model. In addition, it can save so much gas that is
currently being spent by taxis to find passengers.
REFERENCES
[1] Jun Xu , Rouhollah Rahmatizadeh, Ladislau Bölöni,
and Damla Turgut, “Real-Time Prediction of Taxi
Demand Using Recurrent Neural Networks”, IEEE
Transaction on Intelligent transport system, vol. 19,
no. 8, pp. 2572-2581,Aug.2018.
[2] Fei Miao, Shuo Han, Shan Lin, Qian Wang, John A.
Stankovic,AbdeltawabHendawi,DeshengZhang,Tain
he and George J. Pappas, "Data-Driven Robust Taxi
Dispatch Under Demand Uncertainties", IEEE
Transactions on Control Systems Technology,vol.27,
no. 1, pp. 175 - 191, Jan.2019.
[3] Mohammad Saiedur Rahaman, Yongli Ren, Margaret
Hamilton and Flora D. Salim, "Wait Time Prediction
for Airport Taxis Using Weighted Nearest Neighbor
Regression", IEEE Access, vol. 6, pp. 74660 - 74672,
Nov.2018.
[4] Desheng Zhang, Tian He, Shan Lin, Sirajum Munirand
John A. Stankovic,"Taxi-Passenger-DemandModeling
Based on Big Data from a Roving Sensor Network",
IEEE Transactions on Big Data, vol. 3, no. 3, pp. 362 -
374, Sept.2017.
[5] Luis Moreira-Matias, JoãoGama,MichelFerreira,João
Mendes-Moreira, and Luis Damas, "Predicting Taxi–
Passenger Demand Using Streaming Data", IEEE
Transactions on Intelligent Transportation Systems,
vol. 14, no. 3, pp. 1393 - 1402, sept.2013.
[6] Biao Leng, Heng Du, Jianyuan Wang, Li Li and Zhang
Xiong, "Analysis of Taxi Drivers’ Behaviors Within a
Battle BetweenTwoTaxiApps",IEEETransactionson
Intelligent Transportation Systems, vol. 17, no. 1, pp.
296 - 300, jan.2016.
[7] Nicholas Jing Yuan, Yu Zheng,LiuhangZhangandXing
Xie, "T-Finder: A Recommender System for Finding
Passengers and Vacant Taxis", IEEE Transactions on
Knowledge and Data Engineering, vol. 25, no. 10, pp.
2390 - 2403, Oct.2013.
[8] Fei Miao, Student Member, Shuo Han, Member, Shan
Lin, John A. Stankovic,DeshengZhang,SirajumMunir,
Hua Huang, Tian He, and George J. Pappas, "Taxi
Dispatch With Real-Time Sensing Data in
Metropolitan Areas: A Receding Horizon Control
Approach", IEEE TransactionsonAutomationScience
and Engineering, vol. 13, no. 2, pp. 463 - 478,
Mar.2016.
[9] P. Santi, G. Resta, M. Szell, S. Sobolevsky,S.H.Strogatz,
and C. Ratti, “Quantifying the benefits of vehicle
pooling with shareability networks,” Proc. Nat. Acad.
Sci. USA, vol. 111, no. 37, pp. 13290–13294, 2014.

IRJET- Prediction of Cab Demand using Machine Learning

More Related Content

What's hot (19)

Similar to IRJET- Prediction of Cab Demand using Machine Learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Prediction of Cab Demand using Machine Learning