My Projects Portfolio

This comprehensive analysis of the film industry examined over 3,000 movies spanning four decades to identify key factors that predict box office success. The project involved cleaning and integrating data from multiple sources using Python (Pandas) and SQL, creating an interactive dashboard that helps stakeholders visualize profitability trends across different dimensions.

This project developed a robust machine learning pipeline to predict credit card defaults using the UCI dataset. Through extensive data preprocessing and feature engineering, the model achieved a 77% ROC-AUC score, significantly improving credit risk assessment capabilities. The dashboard created enables financial institutions to visualize customer risk segments and implement targeted intervention strategies.

This healthcare analytics project integrated data from multiple sources to create a unified dataset of over 100,000 patient records. Using advanced statistical analysis and machine learning techniques, the project identified significant predictors of hospital readmission and developed models to predict 30-day readmissions for diabetic patients. The interactive dashboard created enables hospital administrators to identify high-risk patients and implement targeted interventions.

This comprehensive retail analytics project examined over 50,000 supermarket transactions to identify sales patterns, seasonal trends, and product correlations. Using advanced machine learning and pattern recognition algorithms, the project derived complex insights and created interactive Tableau dashboards showcasing key performance metrics. The actionable business recommendations developed could increase overall sales by 23% through targeted merchandising strategies.

This e-commerce analytics project analyzed over 100,000 customer transactions across 32,000+ products to identify distinct customer segments and purchasing patterns. The interactive Power BI dashboard created revealed 4 high-value customer segments accounting for 70% of revenue. The targeted retention strategies developed have the potential to increase repeat purchase rate from 3.4% to 5.2%, resulting in $2.4M-$2.9M in additional annual revenue.

This computer vision project developed a convolutional neural network (CNN) to automatically detect and classify plant diseases from leaf images. The model was trained on the PlantVillage dataset containing 54,000 images across 38 disease classes and achieved 94% accuracy on unseen test data. The deployment included a web application interface allowing farmers to upload images and receive instant disease diagnoses, potentially saving crops worth thousands of dollars through early intervention.

Developed a neural network achieving 94% accuracy in pneumonia classification from X-ray images. Implemented transfer learning with MobileNetV2 architecture optimized for resource constraints and 3x data augmentation to overcome limited training data. Engineered custom data augmentation techniques that improved model generalization by 11%. Created an interactive web application with model explainability features using Streamlit.

This project presents a comprehensive supply chain optimization system designed to help small businesses make data-driven inventory decisions. While large enterprises often have sophisticated inventory management systems, small businesses frequently rely on intuition and basic heuristics, leading to inefficiencies and unnecessary costs. This solution bridges that gap by providing advanced analytics capabilities without requiring specialized knowledge.

This telecommunications analytics project leveraged customer data to build a predictive model for identifying customers at risk of churning. Through extensive feature engineering and ensemble learning techniques, the model achieved 91% accuracy in predicting customer churn. The SHAP analysis revealed key factors driving customer attrition, enabling targeted retention campaigns that could save the company an estimated $3.2M annually.

This natural language processing project analyzed over 500,000 product reviews to classify sentiment and extract key product attributes mentioned by customers. Using BERT fine-tuning and topic modeling techniques, the system achieved 89% accuracy in sentiment classification and successfully identified trending product issues and strengths. The interactive dashboard created enables product managers to track sentiment trends and prioritize product improvements based on customer feedback.