PROJECTS

This project, which predicts fraud in financial transactions, aims to identify the best approaches by comparing the performance of different machine learning models. It includes data preprocessing using Python, balancing imbalanced classes with SMOTE, model training, and evaluation. It also provides a Streamlit interface that allows users to upload their own data and perform interactive analysis.

• Comparing the performance of multiple machine learning algorithms on balanced data with SMOTE.
• With the Streamlit interface, users can upload their data and visualize the results of their selected models.

This project segments e-commerce customers into behavior-based segments using RFM analysis and rule-based classification. The segmentation enabled the personalization of marketing strategies.

• 7 customer segments were defined based on RFM scores, and an automated segmentation pipeline was created.
• Cross-selling campaign conversion rates increased by 15%.

A machine learning-based classification model was developed to predict the churn (customer loss) probability of telecom subscribers. Feature engineering and hyperparameter optimization were used.

• An 8% improvement was achieved in accuracy and AUC scores compared to baseline models.
• A monitoring dashboard was created for the early detection of potential churn customers.

A hybrid recommendation system that combines market basket analysis (ARL) and collaborative filtering techniques for an e-commerce platform. It enhanced both product diversity and personalization.

• The diversity of recommendations increased by 20% through basket-based association rules and user-based filtering.
• User engagement and average basket size were significantly improved.