My customer segmentation project analyzed 541,909 retail transactions and used RFM features to group customers into four segments. The results highlighted VIP and at-risk customers for targeted marketing and retention.
For credit risk, I compared classification models on 466,285 loan records and used SMOTE to address class imbalance. Tuning the decision threshold raised default recall to about 85%.
I analyzed employee attrition data and tuned a Gradient Boosting model, reaching about 84% accuracy and 0.80 ROC-AUC. The analysis identified key attrition drivers and grouped employees into low, medium, and high risk.
My product recommendation project processed 32.4 million order records and compared several recommendation approaches. The co-occurrence model scored best, with Precision@5 of 0.40 and Recall@5 of 0.50; I also developed HTML pages for the AI Course Generator application during my internship at Clistas IT Solutions and Services.

