E-Commerce Revenue Prediction Model

Author

Druvitha H K MSc Student – University of Leeds

Project Overview

This project analyses customer behaviour and marketing factors to predict order-level revenue for an e-commerce fragrance retailer called SweetAroma.

Using transactional order data from July 2025, the project identifies key drivers of revenue and builds a predictive model that estimates the expected value of future customer orders.

The analysis follows the CRISP-DM data science methodology including business understanding, data preparation, modelling and evaluation.

Technologies Used

• R Programming
• Linear Regression
• Data Cleaning
• Exploratory Data Analysis
• Predictive Modelling
• Business Analytics

Dataset

The dataset contains 10,000 e-commerce orders with the following variables:

• revenue – order value in GBP
• ad_channel – marketing channel
• number_past_order – number of previous purchases
• past_spend – historical spending
• time_web – time spent on website
• voucher – discount usage

Project Workflow

1. Data Understanding

Exploratory analysis was conducted to understand revenue distribution and customer behaviour patterns.

2. Data Preparation

• Converted categorical variables into factors
• Handled missing values using listwise deletion
• Cleaned dataset reduced to 9,419 complete observations

3. Exploratory Data Analysis

Visualisations were created to analyse:

• revenue distribution
• historical customer spending
• advertising channel distribution
• voucher usage patterns

4. Predictive Modelling

A multiple linear regression model was built to estimate order revenue using behavioural and marketing variables.

Model formula:

Revenue ~ Ad Channel + Past Orders + Past Spend + Website Time + Voucher

5. Model Evaluation

Model performance was evaluated using:

• R²
• Adjusted R²
• Root Mean Square Error (RMSE)
• Mean Absolute Error (MAE)

Residual analysis was also performed to validate model assumptions.

6. Revenue Prediction

The final model was applied to predict revenues for 20 new customer orders.

Key Business Insights

• Customers with higher past spending tend to generate higher order revenue.
• Marketing channels influence order value.
• Website engagement time is associated with increased spending.
• Voucher usage can affect purchasing behaviour.

These insights can help SweetAroma target high-value customers and optimise marketing campaigns.

Skills Demonstrated

• Business analytics
• Data preprocessing
• Statistical modelling
• Predictive analytics
• Data visualisation
• Customer behaviour analysis

Future Improvements

• Test advanced machine learning models
• Deploy model as a business decision tool
• Integrate real-time marketing data

License

MIT License