Prediction of Bike Rental 

Abhishek singh pushkar 

 
 1 
 

Contents 
1.Introduction 
1. Problem Statement 
2. Data 
 
2. Methodology 
1. Data Preparation 
2. Exploratory Data analysis 
I. Distribution of categorical variable with target variable 
2. Distribution of continuous variable with target variable 
3. Distribution of continuous variables. 
3. Missing Value Analysis 
4. Outlier Detection 
5. Outlier Removal 
6. Feature Selection 
7. Sampling 
8. Modelling 
9. Evaluation 
 
3. Conclusion 
 
4. Appendix A 
 
 
 
 
 
 2 
 

Introduction 
Problem Statement 
The  objective  of  this  Case  is  to  Predication  of  bike  rental  count  on  daily  based  on  the 
environmental  and  seasonal  settings.  Using  statistics  we  need  to  read  the  data  and 
develop  a  machine-learning  algorithm  to  predict  the  number  of  bikes  hired  based 
according to the given condition. 
 

Data 
The data given has 731 rows and 16 columns which includes 
 

 
 
Here  our  target  is  the  cnt  variable  which  is  the  count  of  no.  of  bikes  hired.  We  can  see 
from  the  above  data  that  cnt  variable is the sum of casual and registered variables. The 
casual  represents  casual  no.  of  users  who  are  random  travellers  and  registered 
represents the no. of users who are registered by the company as bike users. 
 3 
 

Methodology 
Data preparation 
We need to change the numerical value to the categorical value of the variable  
season: Season  
1: Spring 
2: Summer 
3: Fall  
4: Winter 
yr: Year  
0: 2011 
1:2012  
holiday:  
0: Working Day 
1: Holiday 
Workingday: 
0: Holiday 
1: Working Day  
weathersit: 
1: Clear, Few clouds, Partly cloudy, Partly cloudy  
2: Mist + Cloudy, Mist + Broken clouds, Mist + Few clouds, Mist  
3: Light Snow, Light Rain + Thunderstorm + Scattered clouds, Light Rain  
4: Heavy Rain + Ice Pellets + Thunderstorm + Mist, Snow + Fog  
 
And changing the variables to proper name to increase the readability of the data 
season : Season 
yr : Year 
mnth : Month 
holiday : Holiday 
weekday : Weekday 
workingday : Working Day 
weathersit : Weather Condition 
temp :Temperature 
atemp : Feeling Temperature 
hum : Humidity 
windspeed : Wind Speed 
casual  : Casual Users 
registered : Registered Users 
cnt  : Count 
 4 
 

Exploratory data analysis 

 
Fig.  Bar  graph  showing  the  number  of  bikes  (Count)  hired  season-wise  based  on 
weather condition. 
 

 
Fig. Bar graph showing the no. of bike hired yearly. 
 5 
 

 
Fig. Bar graph showing no. of bikes hired month wise. 
 

 
Fig.  Bar  graph  showing  no.  of  bikes  hired  weekly with separation based on weather the 
day was a holiday or working day. 
 
It  is  observed  from  these  graphs  that  most  of  the  bikes  were  hired  in  the  fall  season 
followed  by  fall.  Users  were  active  when  the  sky  was  clear  or  with  little  rain.  In  2012 
there  are  more  numbers  of  users,  that  is  because  of  the  increase  in  the  number  of 
registered users. 
 
 6 
 

Distribution of Continuous variable with target variable 

 
 
Distribution of continuous variables

 
 7 
 

Missing Value Analysis 

 
Here we can see the outliers in Humidity, Wind Speed and Casual Users. 
 
Outlier Removal 
Outliers can be removed Inter Quarentile Range. IQR is calculated with min and max 
value of a variable. Any value outside the min and max is considered an outlier.  
 8 
 

Feature Selection 

 
Here we need to select the feature which is valuable for the prediction task. Any 
variable with high collinearity should be discarded. From the above image, we can see 
that Temperature and Feeling Temperature are highly correlated so dropping Feeling 
Temperature. Also, Registered users is highly correlated and casual users is of no use so 
dropping these two variables. 
 
Sampling 
Split the dataset into 80 per cent training data and 20 per cent test data. 
 
 
 
 
 9 
 

Modeling 
 
The target variable is continuous therefore the models should be regression type. 
 
Tested with three algorithms: 
1. Linear Regression 
2. Decision Tree Regressor 
3. Random Forest Regressor 
 
 
Evaluation 
 
For score evaluation below are calculated 
1. Mean Absolute Error 
2. Mean Squared Error 
3. Mean Absolute Percentage Error 
4. R squared value 
 
And the value of these scores are: 
1. Mean Absolute Error 
a. LR : 24.75 
b. DTR : 24.42 
c. RFR : 21.72 
 
2. Mean Squared Error 
a. LR : 818.01 
b. DTR : 867.84 
c. RFR : 667.70 
 
3. Mean Absolute Percentage Error 
a. LR : 22.05 
b. DTR : 23.09 
c. RFR : 20.00 
 
4. R Squared value 
a. LR : 0.76 
b. DTR : 0.73 
c. RFR : 0.82 
 
 
 10 
 

Conclusion 
Calculation  of  the  different  scores  tells  us  how  often  an  algorithm  can  predict  future 
cases. These are the definition of how these scores evaluate on the data provided. 
MAE  measures  the  average  magnitude  of  the  errors  in  a  set  of  predictions,  without 
considering  their  direction.  It’s  the  average  over  the  test  sample  of  the  absolute 
differences  between  prediction  and  actual  observation  where  all  individual  differences 
have equal weight. 
MSE is a quadratic scoring rule that also measures the average magnitude of the error. 
MAPE  is  a  measure of prediction accuracy It usually expresses accuracy as a percentage 
and is defined by the formula. 
R-squared  is  a  statistical measure of how close the data are to the fitted regression line. 
It  is  also  known  as  the  coefficient  of  determination,  or  the  coefficient  of  multiple 
determination for multiple regression. 
So  considering  these  different  scores  of  evaluation  and  comparing  them  among 
different algorithms it concludes that Random forest is the best fit for the dataset. 
 

Appendix A 
Distribution of categorical variable with the target variable 
 11 
 

 
Fig.  Bar  graph  showing  the  number  of  bikes  (Count)  hired  season-wise  based  on 
weather condition. 
 

 
Fig. Bar graph showing the no. of bike hired yearly. 
 12 
 

 
Fig. Bar graph showing no. of bikes hired month wise. 
 

 
Fig.  Bar  graph  showing  no.  of  bikes  hired  weekly with separation based on weather the 
day was a holiday or working day. 
   
 13 
 

Distribution of Continuous variable with target variable 

 
 
Distribution of continuous variables

 
 14 
 

Outlier detection 

 
Fig. Boxplots for outlier detection   
 15 
 

 
Feature Selection 

 
 
Fig. Correlation matrix