Customer-Behavior-Analysis and Recommendation

This project includes two parts: user churn rate prediction and song recommendation

Predicted music app user churn rates

1. Performed ETL to clean and feature engineer unstructured app user behavior events data

2. Built ensemble models e.g., random forest

3. Conducted cost-benefit analysis for best retention strategy

Built a song recommendation system employing collaborative filtering

1. Cleaned 14G unstructured log files from smartphones (1.6 billion rows) utilizing text mining

2. Constructed implicit rating scores with a user ad-hoc quantile algorithm and rectified cold start issues by using content-based and popularity-based methods