Electric Vehicle Range Prediction -

Regression Analysis

(By Kothapalli Jayanth (Team Lead), Yashwanth

Goduguchintha, Avesh, Laxman Kumar Busetty and
Prathamesh)

1
1.Fermi estimation:
Fermi estimation, also known as order-of-magnitude estimation, is a
problem-solving technique used to make rough calculations and
approximate solutions based on reasonable assumptions. To perform a
Fermi estimation for an electric vehicle market project, we need to
break down the problem statement into smaller components. Here's an
example breakdown:

Total global population:

 Estimate the current global population, e.g., 7.9 billion.

Percentage of people who can drive:
Assume that approximately 70% of the global population is of
driving age and can drive.

Average household size:

Assume an average household size of 3 people.

Percentage of households that own a vehicle:
Estimate the percentage of households that own a vehicle, e.g., 60%.

Average vehicle ownership per household:

Assume an average of 1.5 vehicles per owning household.

Market penetration of electric vehicles:
Estimate the percentage of vehicles in the market that are electric,
e.g., 10%.

Vehicle lifespan:

2
Determine the average lifespan of vehicles, e.g., 10 years.

Replacement rate:

Assume that each vehicle is replaced with a new one after its lifespan
ends.

Electric vehicle market growth rate:

Estimate the annual growth rate of the electric vehicle market, e.g.,
20%.

Market share of electric vehicles:

Determine the percentage of the overall market that electric vehicles

hold, based on their growth rate and market penetration.
Total number of electric vehicles in the market:

Calculate the cumulative number of electric vehicles based on the

market share and total number of vehicles.
Charging infrastructure:

Estimate the number of charging stations required to support the

growing electric vehicle market.
Investment in charging infrastructure:

Assess the estimated cost of building the necessary charging

infrastructure.
Environmental impact:

Analyze the potential reduction in carbon emissions based on the

adoption of electric vehicles.
3
By breaking down the problem into these components, we can make
reasonable estimations and assumptions to arrive at approximate
figures related to the electric vehicle market project

2. Data Collection:

https://www.tesla.com/ownersmanual/modely/en_kr/GUID-4AC32116-
979A-4146-A935-F41F8551AFE6.html

https://iq.opengenus.org/advantages-and-disadvantages-of-linear-
regression/

https://www.statology.org/linear-regression-assumptions/

Data set link: https://www.kaggle.com/datasets/geoffnel/evs-one-

electric-vehicle-dataset

Data Preprocessing

Libraries

0.1.1 L
o
aimport pandas as pd
dimport numpy as np
iimport matplotlib.pyplot as plt
n%matplotlib inline
gimport seaborn as sns
import plotly.express as px
4
 from scipy.stats import norm
import warnings
warnings.filterwarnings('ignore')

0.1.2 Reading csv

[2]: data=pd.read_csv('C:/Users/divya/Downloads/data_new.csv',usecols=lambda col:␣

↪col not in ['Unnamed: 0'])

df=data.copy()
df.head()

[2]: Brand Model AccelSec TopSpeed_KmH \

0 Tesla Model 3 Long Range Dual Motor 4.6 233
1 Volkswagen ID.3 Pure 10.0 160
2 Polestar 2 4.7 210
3 BMW iX3 6.8 180
4 Honda e 9.5 145

Range_Km Efficiency_WhKm FastCharge_KmH RapidCharge PowerTrain \

0 450 161 940 Yes AWD
1 270 167 250 No RWD
2 400 181 620 Yes AWD
3 360 206 560 Yes RWD
4 170 168 190 Yes RWD

PlugType BodyStyle Segment Seats Price

0 Type 2 CCS Sedan D 5 5002354.20
1 Type 2 CCS Hatchback C 5 2704950.00
2 Type 2 CCS Liftback D 5 5088912.60
3 Type 2 CCS SUV D 5 6134826.60
4 Type 2 CCS Hatchback B 4 2975174.51

0.1.3 Initial Review

[3]: df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 103 entries, 0 to 102
Data columns (total 14 columns):
# Column Non-Null Count Dtype

0 Brand 103 non-null object

1 Model 103 non-null object
2 AccelSec 103 non-null float64
3 TopSpeed_KmH 103 non-null int64
4 Range_Km 103 non-null int64
5 Efficiency_WhKm 103 non-null int64

5
 6 FastCharge_KmH 103 non-null int64
7 RapidCharge 103 non-null object
8 PowerTrain 103 non-null object
9 PlugType 103 non-null object
10 BodyStyle 103 non-null object
11 Segment 103 non-null object
12 Seats 103 non-null int64
13 Price 103 non-null float64
dtypes: float64(2), int64(5), object(7)
memory usage: 11.4+ KB

[4]: df.describe()

[4]: AccelSec TopSpeed_KmH Range_Km Efficiency_WhKm FastCharge_KmH \

count 103.000000 103.000000 103.000000 103.000000 103.000000
mean 7.396117 179.194175 338.786408 189.165049 444.271845
std 3.017430 43.573030 126.014444 29.566839 203.949253
min 2.100000 123.000000 95.000000 104.000000 170.000000
25% 5.100000 150.000000 250.000000 168.000000 260.000000
50% 7.300000 160.000000 340.000000 180.000000 440.000000
75% 9.000000 200.000000 400.000000 203.000000 555.000000
max 22.400000 410.000000 970.000000 273.000000 940.000000

Seats Price
count 103.000000 1.030000e+02
mean 4.883495 5.033056e+06
std 0.795834 3.077267e+06
min 2.000000 1.814931e+06
25% 5.000000 3.104336e+06
50% 5.000000 4.057425e+06
75% 5.000000 5.860725e+06
max 7.000000 1.938548e+07

[5]: df = df.astype({'Brand': str, 'Model': str})

df['Model_Brand']=df['Brand']+df['Model']
df.drop(['Brand','Model'],axis=1,inplace=True)

[6]: df.head()

[6]: AccelSec TopSpeed_KmH Range_Km Efficiency_WhKm FastCharge_KmH \

0 4.6 233 450 161 940
1 10.0 160 270 167 250
2 4.7 210 400 181 620
3 6.8 180 360 206 560
4 9.5 145 170 168 190

RapidCharge PowerTrain PlugType BodyStyle Segment Seats Price \

6
 0 Yes AWD Type 2 CCS Sedan D 5 5002354.20
1 No RWD Type 2 CCS Hatchback C 5 2704950.00
2 Yes AWD Type 2 CCS Liftback D 5 5088912.60
3 Yes RWD Type 2 CCS SUV D 5 6134826.60
4 Yes RWD Type 2 CCS Hatchback B 4 2975174.51

Model_Brand
0 Tesla Model 3 Long Range Dual Motor
1 Volkswagen ID.3 Pure
2 Polestar 2
3 BMW iX3
4 Honda e

0.1.4 Exploration

[7]: cat_column=['RapidCharge','PowerTrain','PlugType','BodyStyle','Segment','Seats']

[8]: for attribute in cat_column:

sns.set(style='whitegrid')
plt.figure(figsize=(8,5))
sns.countplot(df[attribute])
plt.xticks(rotation=90)
plt.title(f'Frequency plot of {attribute}')
plt.show()

7
8
9
10
11
 Interpretation
• There are a greater amount of rapid charge EV vehicles.
• The AWD power train of vehicles are in large number.
• Type 2 CCS plug type vehicles are larger in number.
• The SUVs are in a large number.
• C(Medium Segment) vehicles are in abundance.
• Most of the vehicles have 5 seats.

[9]: df['Seats']=df['Seats'].astype('str')

[10]: df['RapidCharge']=df['RapidCharge'].map({'Yes':1,'No':0})

[11]: # high value preferred columns

num_column=['TopSpeed_KmH','Range_Km','FastCharge_KmH']

num_df=df[num_column]
num_df.set_index(keys=df.Model_Brand,inplace=True)

[12]: def plot_num_att(att):

top_val = num_df[att].sort_values(ascending=False)[:10]
sns.set(style='whitegrid')
plt.figure(figsize=(8,5))
sns.barplot(x=top_val.index, y=top_val)
plt.xticks(rotation=90)
plt.title(f'Bar plot of {att}')
plt.show()

[13]: for att in num_column:

plot_num_att(att)

12
13
14
 Interpretation
• Tesla Roadster is the fastest among all.
• Again Tesla Roadster has the highest range/km ratio, followed by tesla cybertruck and lucid
air.
• The vehicles Tesla model 3 long range, tesla y long range and tesla roadster have the highest
fast charge efficiency.

[14]: df.columns

[14]: Index(['AccelSec', 'TopSpeed_KmH', 'Range_Km', 'Efficiency_WhKm',

'FastCharge_KmH', 'RapidCharge', 'PowerTrain', 'PlugType', 'BodyStyle',

15
 'Segment', 'Seats', 'Price', 'Model_Brand'],
dtype='object')

[15]: # low values preferred columns

num_column=['AccelSec','Efficiency_WhKm','Price']
num_df=df[num_column]
num_df.set_index(keys=df.Model_Brand,inplace=True)

[16]: def plot_num_att(att):

top_val = num_df[att].sort_values()[:10]
sns.set(style='whitegrid')
plt.figure(figsize=(8,5))
sns.barplot(x=top_val.index, y=top_val)
plt.xticks(rotation=90)
plt.title(f'Bar plot of {att}')
plt.show()

[17]: for att in num_column:

plot_num_att(att)

16
17
18
 Interpretation
• Tesla Roadster has the lowest acclereation second among all.
• Lightyear One has the lowest efficiency WhKm, which means it is most efficient.
• The vehicles seat mini, smart eq, volkswagon e-up and smart eq four have the lowest prices.

0.1 Checking Outliers

[18]: num_attributes = ['AccelSec', 'Efficiency_WhKm', 'Price', 'TopSpeed_KmH',␣

↪'Range_Km', 'FastCharge_KmH']

sns.set(style='whitegrid')
fig, axs = plt.subplots(nrows=2, ncols=3, figsize=(10, 8))

colors = sns.color_palette('muted', n_colors=len(num_attributes))

for i, attribute in enumerate(num_attributes):

19
 row = i // 3
col = i % 3
sns.boxplot( y='value', data=pd.melt(df[[attribute]]), ax=axs[row, col],␣
↪color=colors[i])

axs[row, col].set_xlabel('')
axs[row, col].set_ylabel(attribute)

fig.suptitle('Box plots of numerical attributes', fontsize=14)

plt.tight_layout()
plt.show()

• Except FastCharge every other attribute has outliers present

[19]: fig = plt.figure(figsize=(8,5))

sns.heatmap(df[num_attributes].corr(),annot=True)
plt.show()

20
 There is a strong positive correlation between: * Price and Fast_charge. * Price and Range_km.
* Price and Top Speed. * Top Speed and Fast Charge. * Top Speed and Range km. * Range_km
and Fast charge.
There is a negative correlation between: * Accel_sec is negatively correlated with every other
attribute.

0.2.1 Multivariate Analysis

[20]: sns.pairplot(df[num_attributes],diag_kind='kde')
plt.show()

21
[21]: def check_skweness(df,columnName):

try:
(mu, sigma) = norm.fit(df[columnName])
except RuntimeError:
(mu,sigma) = norm.fit(df[columnName].dropna())
print("Mu {} : {}, Sigma {} : {}".format(
columnName.upper(), mu, columnName.upper(), sigma))
plt.figure(figsize=(20,10))
sns.distplot(df[columnName], fit=norm, color="orange")
plt.title(columnName.upper() +
" Distplot", color="black")
plt.show()

22
[22]: for columns in df[num_attributes].columns:
check_skweness(df[num_attributes],columns)

Mu ACCELSEC : 7.39611650485437, Sigma ACCELSEC : 3.0027470377696464

Mu EFFICIENCY_WHKM : 189.16504854368932, Sigma EFFICIENCY_WHKM :

29.42296081389322

Mu PRICE : 5033055.821747573, Sigma PRICE : 3062291.940017147

23
Mu TOPSPEED_KMH : 179.19417475728156, Sigma TOPSPEED_KMH : 43.36099501160743

Mu RANGE_KM : 338.7864077669903, Sigma RANGE_KM : 125.40123181797216

24
Mu FASTCHARGE_KMH : 444.2718446601942, Sigma FASTCHARGE_KMH : 202.95679363121414

• The numerical attributes are mostly normally distributed.

25
 0.2 Cluster Analysis

[23]: from sklearn.pipeline import Pipeline

from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.cluster import KMeans

[24]: cat_column=['PowerTrain', 'PlugType', 'BodyStyle', 'Segment', 'Seats']

0.3.1 Determining optimal cluster solution

0.3.2 K Means clustering

[25]: elbow = dict()

preprocessor = ColumnTransformer(
transformers=[
('num', StandardScaler(), [col for col in num_attributes]),
('cat', OneHotEncoder(), [col for col in cat_column])
])

for i in range(2,8):

pipeline = Pipeline([
('preprocessor', preprocessor),
('estimator', KMeans(n_clusters=i))

])
clusters = pipeline.fit_predict(df)
kmeans_estimator = pipeline.named_steps['estimator']
elbow[i] = kmeans_estimator.inertia_

#Plotting Elbow plot

plt.plot(*zip(*elbow.items()), marker='o', linestyle='--')
plt.title('Elbow Method for Optimal k')
plt.ylabel('Inertia')
plt.xlabel('Clusters');

26
 In this elbow plot k=4 seems the optimal solution

Choosing n_clusters=4 as optimal solution

[26]: pipeline = Pipeline([
('preprocessor', preprocessor),
('estimator', KMeans(n_clusters=4))
])

df['Labels'] = pipeline.fit_predict(df)

[27]: df['RapidCharge']=df['RapidCharge'].map({1:'Yes',0:'No'})

0.3 Studying Segments

[28]: df['Seats']=df['Seats'].astype('int')
df.groupby('Labels').mean()

[28]: AccelSec TopSpeed_KmH Range_Km Efficiency_WhKm FastCharge_KmH \

Labels

27
 0 6.216667 185.416667 380.694444 211.194444 510.555556
1 12.681818 132.545455 154.090909 173.727273 189.090909
2 8.660526 153.447368 278.815789 170.894737 313.684211
3 3.855556 249.611111 494.444444 193.111111 743.333333

Seats Price
Labels
0 5.055556 5.361855e+06
1 4.000000 2.451013e+06
2 4.868421 3.184689e+06
3 5.111111 9.855480e+06

The previous attribute information provides that: * Lower values AccelSec are generally preferred
as they indicate faster acceleration. * Higher values TopSpeed_KmH are generally preferred as
they indicate greater speed. * Higher values Range_Km are generally preferred as they indicate
greater range. * Lower values Efficiency_WhKm are generally preferred as they indicate better
energy efficiency. * Higher values FastCharge_KmH are generally preferred as they indicate faster
charging. * Lower values Price are generally preferred as they indicate greater affordability. *
Higher values Seats are generally preferred as they indicate more passenger capacity.

Renaming labels
[29]: df['Labels']=df['Labels'].map({1:'High1-Range',2:'High2-Range',3:'Mid-Range',0:
↪'Low-Range'})

0.4 Visualising Clusters

[30]: from sklearn.manifold import TSNE

def Visualise_cluster(data):

model = TSNE(random_state=1)
transformed = model.fit_transform(preprocessor.fit_transform(data))

plt.title('Flattened Graph of 4 Clusters')

sns.scatterplot(x=transformed[:,0], y=transformed[:,1], hue=data['Labels'],␣
↪style=data['Labels'], palette="Set1")

Visualise_cluster(df)

28
 Interpretation :
The market is majorly dominated by mid range and a mix of high and mid car models, whereas
the low end and the high end vehicles are present in fewer number.

0.5.1 Ward Linkage

[31]: df1=data.copy()
df1 = df1.astype({'Brand': str, 'Model': str})
df1['Model_Brand']=df1['Brand']+df1['Model']
df1.drop(['Brand','Model'],axis=1,inplace=True)
df1['Seats']=df1['Seats'].astype('str')
df1['RapidCharge']=df1['RapidCharge'].map({'Yes':1,'No':0})

[32]: from sklearn.cluster import AgglomerativeClustering

from scipy.cluster.hierarchy import dendrogram

[33]: ward = AgglomerativeClustering(distance_threshold=0, n_clusters=None,␣

↪linkage='ward')

clusters_customers = ward.fit_predict(preprocessor.fit_transform(df1))

29
[34]: def plot_dendrogram(model, **kwargs):
# create linkage matrix and then plot the dendrogram

# create the counts of samples under each node

counts = np.zeros(model.children_.shape[0])
n_samples = len(model.labels_)
for i, merge in enumerate(model.children_):
current_count = 0
for child_idx in merge:
if child_idx < n_samples:
current_count += 1 # leaf node
else:
current_count += counts[child_idx - n_samples]
counts[i] = current_count

linkage_matrix = np.column_stack([model.children_, model.distances_,␣

↪counts]).astype(float)

# plot the corresponding dendrogram

dendrogram(linkage_matrix, **kwargs)

[35]: plt.title('Dendrogram with Ward Linkage')

plot_dendrogram(ward, truncate_mode='lastp')

30
 • According to the dendrogram 2 cluster solution is appropriate, in case of Ward’s Linkage

[36]: pipeline1 = Pipeline([

('preprocessor', preprocessor),
('estimator', AgglomerativeClustering(distance_threshold=None,␣
↪n_clusters=2, linkage='ward'))

])

df1['Labels']=pipeline1.fit_predict(df1)

[37]: df1['RapidCharge']=df1['RapidCharge'].map({1:'Yes',0:'No'})

0.5 Studying Segments

[38]: df1['Seats']=df1['Seats'].astype('int')
df1.groupby('Labels').mean()

[38] : AccelSec TopSpeed_KmH Range_Km Efficiency_WhKm FastCharge_KmH \

Labels
0 5.102083 213.083333 419.375000 205.958333 597.291667

31
 1 9.398182 149.618182 268.454545 174.509091 310.727273

Seats Price
Labels
0 5.083333 7.169506e+06
1 4.709091 3.168518e+06

• The above labels suggest that label 0 are high end vehicles.
• The label 1 has vehicles that fall under mid-end section vehicles.

Renaming Labels
[39]: df1['Labels']=df1['Labels'].map({1:'Mid-Range',0:'High-Range'})

0.6 Visualising Clusters

[40]: Visualise_cluster(df1)

• For a two cluster solution the market is evenly distributed for High and Mid range

32
Segment Extraction:
1 1. Feature Engineering: If needed, perform feature engineering
. techniques to preprocess the input features. This may include data
normalization, handling missing values, encoding categorical
F variables, and transforming features for better representation.
e
a 1. Train-Test Split: Split the dataset into training and testing sets.
t
u The training set is used to train the Gradient Boosting Regressor
r model, while the testing set is used to evaluate its performance
e
and generalization ability.
E 2. Model Training: Fit the Gradient Boosting Regressor to the
n
g training data. The model learns to iteratively improve the
i predictions by minimizing the residuals (errors) of the previous
n
e models in the ensemble. Gradient Boosting combines multiple
e decision trees, where each subsequent tree corrects the errors
r
made by the previous trees.
i
n
3. Hyperparameter Tuning: Tune the hyperparameters of the
g
: Gradient Boosting Regressor model to optimize its performance.
This may involve adjusting parameters such as the learning
I
f rate, number of estimators (trees), maximum depth of the trees,
and minimum samples required for a leaf node.
n
e 4. Model Evaluation: Evaluate the trained model using the testing
e
set. Calculate relevant regression metrics such as mean squared
d
e error (MSE), mean absolute error (MAE), or R-squared to assess
d the model's accuracy and reliability in predicting the electric
,
vehicle range.
p
e 5. Segment Extraction: Once the Gradient Boosting Regressor
r model is trained and validated, it can be used to extract
f
segments based on the importance of the input features. The
33
m to the relative significance of different features in predicting the
o range, thereby identifying the segments where specific features
d play a more crucial role.
e
6, Interpretation and Analysis: Analyze the results and
l interpretations from the model to understand the distinct

segments within the electric vehicle range prediction. This can

c help identify patterns and relationships that differentiate various
a segments, such as the impact of battery capacity, driving behavior,
n or environmental conditions on the range.

p
ML Models:-
r
o 1. In the context of electric vehicle range prediction using regression
analysis, several ML techniques can be utilized to extract segments.
v
Regression analysis aims to establish a relationship between input
i variables (features) and the target variable (electric vehicle range) to
d make predictions. Here are some ML techniques commonly used for
segment extraction in electric vehicle range prediction using regression
e analysis:

i 1. Linear Regression: Linear regression is a fundamental technique

n used to model the relationship between input features and the
s target variable. It assumes a linear relationship between the
i features and the range. Through the estimation of coefficients,
g it determines the weight or importance of each feature in
h predicting the range.
t
2. 2. Ridge Regression: Ridge reRidge regression: is a
s
regularization technique that helps prevent overfitting in
regression models. It adds a penalty term to the linear
i
regression objective function, which encourages smaller
n
coefficient values and reduces the impact of irrelevant or noisy
34
f useful when dealing with high-dimensional data or when
e multicollinearity is present among the input features.
a
t 3. Lasso Regression: Lasso regression is another regularization
u technique that introduces a different penalty term. It not only
r shrinks coefficient values but also performs feature selection by
e driving some coefficients to exactly zero. Lasso regression is
s effective when there is a large number of features, and feature
. selection is desired to identify the most relevant predictors for
range prediction.
R
i
d 4. Support Vector Regression (SVR): SVR is a regression technique
g based on support vector machines. It finds an optimal
e hyperplane that maximizes the margin while still fitting the
training data within a certain margin of error (epsilon). SVR can
r handle non-linear relationships by using kernel functions and is
e robust to outliers.
g
r
5. Gradient Boosting Regressor is a machine learning technique
e that can be used for segment extraction in electric vehicle
s range prediction through regression analysis. Gradient
s Boosting is an ensemble method that combines multiple weak
i learners (decision trees) to create a strong predictive model.
Here's how Gradient Boosting RegrRregressor can be used for
o
segment extraction.
n

 Profiling :
i  process of creating user profiles or individualized representations
s based on their specific characteristics, preferences, and historical data.
35
T model to provide personalized range predictions for different users.
h Here's how profiling can be incorporated into an ML model for electric
e vehicle range prediction:
s  Data Collection: Collect data from individual electric vehicle
e
users, including their driving patterns, charging habits,
p historical range information, and other relevant features. This
r
o data serves as the basis for creating user profiles.
f
 User Segmentation: Analyze the collected data to identify
i
l different user segments based on their driving behavior,
e charging patterns, and other relevant factors. This
s
segmentation helps group users with similar characteristics
h together, allowing for more targeted profiling.
e
l  Profile Creation: For each user segment, create individual
p
profiles that capture their specific characteristics and
t preferences. The profiles can include information such as
a average daily distance traveled, charging frequency, preferred
i
l charging locations, and other user-specific details.
o
r  Feature Extraction: Extract relevant features from the user
profiles that are likely to impact range prediction. These
t
h features can include driving distance, time of day, weather
e conditions during the commute, charging duration, and any
other personalized attributes identified during profiling.
r
e  Model Training: Use the extracted features from the user
g
r profiles along with the corresponding range values to train the
e regression model. The model learns the relationship between
s
s the personalized features and the range, enabling it to provide
i individualized range predictions.
o
n
 Code Link:
36
 C AveshBhati7/Market-Segementation-EV-cars/tree/main
o
d
e

L
i
n
k
:
h
t
t
p
s
:
/
/
g
i
t
h
u
b
.
c
o
m
/
37