Market Basket Analysis - Machine Learning Project / January, 2023 — March, 2025

Collected and processed over 100,000+ retail transaction records to identify purchasing patterns and frequent item combinations using association rule mining. Generated data cleaning, removing duplicates, handling missing values, and standardizing product names to improve data quality by 20%. Engineered meaningful features including one-hot encoded item matrices, time-based patterns, and customer segments to enhance model insights. Conducted exploratory data analysis (EDA) with 15+ visualizations to reveal high-lift product pairs and seasonality trends informing business decisions. Implemented the Apriori algorithm to generate interpretable association rules, enabling strategic product bundling, targeted marketing, and layout optimization. Delivered actionable insights that improved cross-selling effectiveness and supported tiered discount strategies, driving a projected 25% increase in bundled product sales.

Flight Delay Prediction - Machine Learning Big Data Project (HADOOP + PYTHON) / January, 2023 — December, 2024

Collected and processed 150,000+ flight records from the Data Expo 2009 airline dataset using Hive and Hadoop across five years (1999, 2000, 2005, 2007, 2008) with 30,000 random samples per year. Analyzed distributed data cleaning in Hive: handled missing values, standardized column formats, and added a derived binary 'Delayed' column based on delay thresholds. Exported and merged datasets using HDFS, Hive, and secure copy (SCP) for final integration into a unified modeling dataset. Engineered relevant features by dropping leakage-prone columns, label encoding categorical variables, and replacing missing values using medians. Trained and evaluated multiple models (Logistic Regression, KNN, Decision Tree, Random Forest, Gradient Boosting) tracking accuracy, precision, and recall. Optimized model performance on a dataset of 1M+ data points by conducting comprehensive hyper-parameter tuning on Decision Tree, Random Forest, and Gradient Boosting algorithms using GridSearchCV. Best model: Tuned Gradient Boosting Classifier with 92.36% accuracy, 94.73% precision, and 91.28% recall on validation data. Applied the final model to test data, generated predictions, and exported results as testpredictionsfinals.csv.