This project presents a comprehensive data mining analysis of sales and customer behavior on the Amazon e-commerce platform, focusing on the Indian market. By leveraging two distinct datasets—a sales report with ~130,000 transactions and a product dataset with over 1,400 products and their reviews—this analysis uncovers actionable insights to drive strategic business decisions.
The primary goal is to transform raw data into a clear understanding of what drives sales, customer satisfaction, and operational efficiency, ultimately providing a blueprint for optimizing marketing, inventory, and customer experience.
This project answers critical business questions through a multi-faceted analytical approach:
- Sales & Revenue Forecasting: What are the key drivers of revenue, and can we predict future sales trends?
- Order Cancellation Prediction: Why do customers cancel orders, and can we identify at-risk orders proactively?
- Product & Customer Segmentation: What distinct groups of products and customers exist, and how can we tailor strategies for them?
- Customer Sentiment Analysis: What is the underlying sentiment in customer reviews, and how does it impact product perception?
- Market Basket Analysis: Are there products that are frequently purchased together?
- Language: Python
- Data Manipulation & Analysis: Pandas, NumPy
- Machine Learning: Scikit-learn, TensorFlow (Keras), Statsmodels
- Data Visualization: Matplotlib, Seaborn
- Environment: Jupyter Notebook
A structured data mining pipeline was followed: Data Cleaning → Exploratory Data Analysis (EDA) → Feature Engineering → Model Development → Evaluation.
Here are the standout findings and their business impact:
-
Price Prediction (Regression): A Neural Network (ANN) was developed that achieved an outstanding R² score of 0.98.
- Business Impact: Enables highly accurate, dynamic pricing strategies and promotional planning.
-
Order Cancellation (Classification): A Random Forest Classifier proved most effective, reaching 95% accuracy.
- Business Impact: Allows the business to identify and intervene with orders likely to be canceled, reducing revenue loss and improving logistics.
-
Using an improved K-Means Clustering model (Silhouette Score: 0.708), products were segmented into three clear, strategic groups:
- 👑 Premium / High-Value Products: High price, high rating.
- 👍 Value-for-Money Products: Low price, but still good ratings.
- ⭐ Bestsellers / Popular Products: Massive review counts, trusted by the public.
- Business Impact: Provides a clear guide for targeted marketing campaigns, inventory management, and personalized customer promotions.
- A Convolutional Neural Network (CNN) was built to analyze customer review text. After addressing significant class imbalance, the final model could effectively identify negative sentiment.
- Business Impact: Offers a powerful tool to automatically gauge customer satisfaction, pinpoint product flaws, and prioritize areas for improvement, directly from the customer's voice.
4. 🔗 Uncovering Hidden Purchase Patterns
- Apriori Association Rule Mining revealed valuable insights, such as:
- “Deeply discounted, high-priced electronics are highly likely to receive a high rating.”
- Business Impact: Justifies strategic promotions on high-value items to boost both sales and customer satisfaction. Informs cross-selling and product bundling strategies.
- A SARIMA time-series model was successfully implemented to forecast daily revenue for the next two months, capturing weekly seasonality.
- Business Impact: Assists in financial planning, resource allocation, and anticipating future inventory needs.
Khai_Pha_Amazon_Sales_Report.ipynb: Main notebook for analyzing the transaction-level sales report. Contains models for cancellation prediction, revenue forecasting, and clustering.Khai_Pha_Amazon_Sales_Dataset.ipynb: Main notebook for analyzing the product and review dataset. Contains models for sentiment analysis, price prediction, and association rule mining.data/: Directory containing the two CSV datasets.