E-commerce Yearly Sales Prediction: App vs Website Focus

A Machine Learning Approach using Linear Regression

---

Introduction

With the growing digital economy, understanding where to focus — mobile app or website — is a critical question for e-commerce businesses. In this project, I used a Kaggle dataset that compares mobile app vs website sales performance and applied Linear Regression to predict Yearly Sales Revenue.

🔗 Dataset link: Kaggle - Focusing on Mobile App or Website

---

📁 Dataset Overview

The dataset provides insights like:

• Customer count

• Time spent on app vs website

• Length of membership

• Session frequency

• Yearly Amount Spent

Our goal was to predict the Yearly Amount Spent using other behavioral features.

---

🛠️ Tools & Libraries Used

pythonCopyEditimport pandas as pd  
import numpy as np  
import matplotlib.pyplot as plt  
import seaborn as sns  
from sklearn.model_selection import train_test_split  
from sklearn.linear_model import LinearRegression  
from sklearn import metrics  
import math  
import pylab  
import scipy.stats as stats

---

📊 Exploratory Data Analysis (EDA)

To start, I used Seaborn and Matplotlib to visualize relationships:

• Heatmap to view feature correlation

• Jointplots to explore how Time on App and Time on Website impact yearly spending

• Pairplots to get a holistic view of data distribution

This revealed strong correlation between Length of Membership and Yearly Amount Spent.

---

⚙️ Model Building

I trained a Linear Regression model to predict Yearly Amount Spent from the other features.

🔁 Steps:

1. Data cleaning & null check

2. Splitting into train/test

3. Model training using LinearRegression() from sklearn

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

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)  
model = LinearRegression()  
model.fit(X_train, y_train)

---

📈 Model Evaluation

After training, I evaluated using:

pythonCopyEdity_pred = model.predict(X_test)  
mae = metrics.mean_absolute_error(y_test, y_pred)  
mse = metrics.mean_squared_error(y_test, y_pred)  
rmse = math.sqrt(mse)

📉 Results:

• Mean Absolute Error (MAE): 7.23

• Mean Squared Error (MSE): 79.81

• Root Mean Squared Error (RMSE): 8.93

---

💡 Key Learnings

• Length of Membership was the most important feature, highlighting customer retention's importance.

• Time on App had a stronger relationship with yearly spending than Time on Website, suggesting companies should consider focusing more on mobile platforms.

• Even a simple Linear Regression model can provide powerful insights when paired with good feature understanding.✅ Conclusion

This project showed how regression can be used not just for prediction but for strategic decision-making—like whether to invest more in mobile or web. I’m excited to keep experimenting and building smarter ML solutions!

Demo Vedio Link

Linkedin_Post