This repository contains the intermediate development stages of the Fashion E-Commerce ML System, including experimentation, feature development, model iterations, database design evolution, exploratory analysis, and incremental project progress.
It serves as a record of the project's development lifecycle and engineering process.
For the complete, cleaned, and finalized implementation of the project, please visit:
🔗 Final Repository: https://github.com/rayyan-merchant/ml-based_Fashion-Ecommerce
This repository may contain:
- Experimental implementations
- Early model versions
- Intermediate datasets and outputs
- Development snapshots
- Database design iterations
- Feature engineering experiments
- Prototypes and testing code
The final repository consolidates the validated components into a cleaner production-ready structure.
- Collaborative Filtering (Implicit ALS / MF)
- Content-Based Similarity (TF-IDF + BERT embeddings)
- Final Hybrid Model: CF + Content + Popularity + Behavior Re-ranking
- BERT-based sentiment classification
- Sentence-Transformer embeddings
- Topic signals + toxicity detection
- Category-level sentiment trends
- Prophet / ARIMA for article & category demand
- Seasonality patterns, trend detection
- Inventory risk alerts
- Clustering (K-Means / GMM / HDBSCAN)
- RFM scores, behavior ratios, category preferences
- Segment-based personalization
- View → Click → Cart → Purchase paths
- Drop-off analysis
- Conversion KPIs + session trajectories
- Trending vs declining products
- Price elasticity
- Customer lifecycle metrics
- Sales & category evolution
1. Raw Data
2. Filtering & Cleaning
3. ETL → ML Datasets (A–F)
4. Preprocessing & Feature Engineering
5. EDA
6. Model Training (CF, CB, Hybrid, NLP, TS, Segmentation)
7. Hyperparameter Tuning
8. Final Models & Evaluations
9. Trend/BI Dashboards
10. API Serving & Integration
- Dataset A: User–Item interactions (CF)
- Dataset B: Article content features
- Dataset C: Customer features (RFM + behavior)
- Dataset D: Time-series sales
- Dataset E: Reviews (sentiment + embeddings)
- Dataset F: Behavioral events + funnels
These datasets drive all downstream ML models.
Languages & Frameworks: Python, Pandas, scikit-learn, PyTorch, Sentence-BERT, Implicit, Prophet/ARIMA
Storage & Processing: PostgreSQL, Parquet, NumPy
Experimentation: MLflow
Optional Serving: FastAPI, Docker
- High-quality hybrid recommender
- Robust sentiment-aware product intelligence
- Accurate demand forecasts
- Actionable customer segments
- Complete behavioral funnel metrics
- Comprehensive BI and trend insights
