Welcome to our solution for the Personalized Solutions Dataquest Challenge, where we built a predictive and recommendation system to enhance user engagement on a financial services platform.

The challenge was to:

• Predict meaningful customer interactions (CLICK, CHECKOUT) from raw behavioral data.
• Recommend relevant financial products to users based on their activity and profile.

1. Dataset Description
2. Data Preparation
3. Exploratory Data Analysis (EDA)
4. Modeling Approaches
5. Evaluation Metrics
6. Recommendation System
7. Feature Importance
8. Wow Moments
9. Conclusion & Next Steps

• Date conversion and missing value handling
• Created rating: DISPLAY = 0, CLICK = 1, CHECKOUT = 2
• Train/Test Split using time-based logic
• Categorical encoding and feature scaling

• Weekly interaction trends visualized for DISPLAY, CLICK, CHECKOUT
• Feature correlation heatmap showing influence of segment, item_type, active_ind

• Logistic Regression (baseline)
• Random Forest
• Bagging Classifier
• Gradient Boosting

• Improved performance using Bagging and Boosting to reduce variance and bias.

We evaluated models using:

• Accuracy
• Precision
• Recall
• F1 Score
• AUC-ROC
• Log Loss
• Precision@10 (for top-N relevance)

• TF-IDF + Cosine Similarity on item descriptions
• Recommended similar items per product

• SVD-based user-item matrix factorization
• Suggested items based on similar users’ behaviors

Key influencing features:

• segment
• active_ind
• item_type

Visualized using:

• Random Forest Feature Importance
• Permutation Importance

• Transformed interaction into numerical ratings for modeling
• Hybrid Recommender System combining content & collaborative filtering
• Ensemble visualizations with trend lines
• Segment-based insights revealed behavioral drivers
• Time-aware test split to simulate real-world predictions
• Precision@K focused on top-N quality

• Accurate interaction prediction with ensemble methods
• Personalized recommendations using hybrid approach

• Add time-series aware modeling
• Improve recommendations for cold-start users
• Automate pipeline for retraining and monitoring

• Simanga Mchunu
• Contact: simacoder@hotmail.com

├── data/
│   └── data.csv
├── notebooks/
│   └── model_training.ipynb
├── src/
│   ├── preprocessing.py
│   ├── modeling.py
│   ├── recommender.py
├── Recommender Presentation.pdf
├── README.md