Using real-world e-commerce data, we’ll build a predictive model that helps understand what drives customer spending.

In this project, we’ll explore:

• How time spent on websites versus mobile apps affects purchasing behavior
• The crucial role of membership duration in customer spending
• Visual analysis of customer engagement patterns
• Building and evaluating a Linear Regression model.

Jump to Code and Explanation

Regression Plot using actual Values

The plot created by sns.lmplot is a regression plot, which is a type of plot used to visualize the relationship between two continuous variables along with a fitted linear regression line.

sns.lmplot(x='Length of Membership',y='Yearly Amount Spent',data=df,scatter_kws={'alpha':0.4})

Following the prediction made by our regression model. In this article we will train the machine learning model and then evaluate it for accuracy.

Complete Code and It’s Explanation

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

# here we imported
# pandas is used to interact with dataframe, clean the data, find means, read files 
# matplotlib and seaborn are used to create visualisation

df=pd.read_csv('ecommerce.csv')

# here we are creating a dataframe df by reading the ecommerce data which is in csv format

df.head()

# it shows first few row, we can also specify how many row to be shows like df.head(10)

df.info()
# It shows information about the data frame like non null count data type, column names 

df.describe()

# It creates statistical summary like count, mean, standard deviation, minimumn value, maximum value 

#Exploratory Data Analysis

sns.jointplot(x='Time on Website',y="Yearly Amount Spent", data=df,alpha=0.5)

# Using Seaborn to create a plot which is a combination of histogram(distribution plot) and scatter plot between two variable 

sns.jointplot(x='Time on App',y="Yearly Amount Spent", data=df,alpha=0.5)

# Using Seaborn to create a plot which is a combination of histogram(distribution plot) and scatter plot between two variable 

sns.pairplot(df, kind='scatter', plot_kws={'alpha': 0.4})
# show various plots which show relationship between various variable in the dataframe
# if the variable is same on y and x axis it shows the histogram
# helped to see all possible relationship 

sns.lmplot(x='Length of Membership',y='Yearly Amount Spent',data=df,scatter_kws={'alpha':0.4})

# scatter plot along with regresstion line , 

from sklearn.model_selection import train_test_split

# this is imported to split the dataset to training and testing set 

X=df[['Avg. Session Length','Time on App','Time on Website','Length of Membership']]
y=df['Yearly Amount Spent']

# creating variable for our model 
# X independent variable
# y dependent variable 
# when we train the model we are trying to predict the y Which is yearly amount spend

X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.3,random_state=42)

# splitting the data for training and testing

X_train
# show the training data

from sklearn.linear_model import LinearRegression

# importing the linear regression model from the scikit-learn

lm=LinearRegression()
# creating an model instance

lm.fit(X_train,y_train)
# Training the LinearRegression model with our data 

lm.coef_

# Show the array of coefficient, how much each variable impact the target value

cdf=pd.DataFrame(lm.coef_,X.columns,columns=['Coef'])
print(cdf)

# Coefficient with higher value has more impact so in this case Lenght of Membership has higher impact on the output

#prediction

predictions=lm.predict(X_test)
predictions

# testing the model with X_test values

# we are testing whether the predictions that the trained model give is similar to the actual

#y_test is the actual y 

# If it is similar then our model is pretty good for using 

sns.scatterplot(x=predictions,y=y_test)
plt.xlabel('Predictions')
plt.title('Evaluation of our LM Model')

The prediction and actual seems to correlating.

Evaluating the Model for Accuracy

from sklearn.metrics import mean_squared_error, mean_absolute_error
import math

# Mean Absolute error check the error in our case it is not bad, mean of yearly amount spend is 500, so error of 8 unit is pretty good


#Average distance between the line and the point of scatter
print('Mean Absolute Error:',mean_absolute_error(y_test,predictions)) # it is like the precictions is off by this much unit

print('Mean Squared Error:',mean_squared_error(y_test,predictions)) # Show larger error 
print('Root Mean Squared Error:',math.sqrt(mean_squared_error(y_test,predictions)))

# These help us know whether our model need more improvement or not.

residuals=y_test-predictions
# Checking How far off each prediction was
# residuals should be random, if not there is some problems or biases

sns.displot(residuals,bins=30,kde=True)
# plot to check the randomness 
#  A good model should have 
# Symmetric distribution
# No unusual pattern
# Bell Shape
# No Multiple peaks

import pylab
import scipy.stats as stats

stats.probplot(residuals,dist="norm",plot=pylab)

# checking whether the residual has normal distribution using Q-Q Plot
# if curved line the data is skewed
# Point far from line - Outlier
# Poitn should close to the diagnal line 
# 

Conclusion

Here we get lm.coef_ (coefficient)

X = customers[[‘Avg. Session Length’, ‘Time on App’, ‘Time on Website’, ‘Length of Membership’]]

Result of lm.coef_
array([25.72425621, 38.59713548, 0.45914788, 61.67473243])

That means Variables which has biggest impact on the Yearly Amount spend is Length of Membership>Time on App>Avg. Session Length>Time on Website

The model in this example has a low prediction error, which means it works well. Using features like Avg. Session Length, Time on App, Time on Website, and Length of Membership, we can reliably predict which variable has the greatest impact on the target which is yearly customer spend.

In this case we find that Length of Membership has the greatest impact on the yearly spend by the customer so we should focus on increasing the membership by making it easy for people to become membership by reducing the membership fees or increasing the perks.