This project explores Amazon product data to understand how pricing, discounts, and other factors influence customer ratings.
The workflow includes:
- Data Cleaning & Preprocessing
- Exploratory Data Analysis (EDA)
- Feature Engineering
- Machine Learning Model Development
- Model Evaluation & Comparison
- Business Insights Extraction
The primary objective is to predict whether a product is likely to receive a high rating based on pricing-related features.
- Analyze Amazon product pricing trends.
- Explore discount patterns across products.
- Identify relationships between pricing and ratings.
- Engineer meaningful predictive features.
- Compare multiple Machine Learning algorithms.
- Predict high-rated products using supervised learning.
| Category | Tools |
|---|---|
| Programming | Python |
| Data Processing | Pandas, NumPy |
| Visualization | Matplotlib, Seaborn |
| Machine Learning | Scikit-Learn |
| Development | Jupyter Notebook |
| Version Control | Git & GitHub |
amazon-product-analysis/
│
├── dataset/
│ └── amazon.csv
│
├── notebook/
│ └── amazon_DA.ipynb
│
├── screenshots/
│ ├── 01_project_overview.png
│ ├── 02_dataset_preview.png
│ ├── 03_data_cleaning_summary.png
│ ├── 04_datset_info.png
│ ├── 05_rating_distribution_graph.png
│ ├── 06_price_distribution.png
│ ├── 07_discount_distribution.png
│ ├── 08_price_vs_rating_analysis.png
│ ├── 09_correlation_heatmap.png
│ ├── 10_model_comparison.png
│ ├── 11_feature_importance.png
│ └── 12_confusion_matrix.png
│
├── report
│ └──analysis_report.pdf
│
└── README.md
The dataset underwent multiple preprocessing steps:
- Missing value inspection
- Data type conversion
- Currency symbol removal
- Percentage formatting cleanup
- Numerical feature extraction
- Data validation
This ensured consistent and reliable inputs for analysis and model training.
Additional features were created to improve model performance:
| Feature | Description |
|---|---|
| Savings | Actual Price − Discounted Price |
| Price Ratio | Discounted Price / Actual Price |
| High Rating | Binary target variable |
These features helped capture pricing behavior more effectively.
The following classification models were trained and evaluated:
- Logistic Regression
- K-Nearest Neighbors (KNN)
- Decision Tree Classifier
- Random Forest Classifier
The models were evaluated using accuracy scores to identify the best-performing approach.
- Most Amazon products maintain ratings above 4.0.
- Large discounts do not always lead to higher ratings.
- Product price has a relatively weak correlation with ratings.
- Pricing-derived features contribute significantly to prediction accuracy.
- Ensemble methods demonstrated stronger predictive performance compared to simpler models.
- Incorporate Natural Language Processing (NLP) on customer reviews.
- Perform sentiment analysis.
- Build a recommendation engine.
- Deploy the model using Streamlit.
- Integrate larger real-world e-commerce datasets.
Rushit Tholiya