Machine Learning Engineer
CurrentConducted Exploratory Data Analysis (EDA) with Presto on an extensively imbalanced credit card dataset, with an imbalance ratio stood at 99.7 to 0.3, simulated authentic real-world scenarios, processed in excess of 12 million transactions, and attained remarkable accuracy in identifying fraudulent transactions.Executed data preprocessing and feature engineering using Apache Spark, reducing the original DataFrame from 24 to 8 columns to optimize model learning capabilities through the creation of new features.Researched and assessed various Machine Learning models, employing systematic testing and validation methodologies to optimize performance metrics and ensure alignment with project specifications.Developed an XGBoostmodel pipeline, integrating hyperparameter tuning and gradient boosting, which resulted in achieving 80% precision, 75% recall, and an MCC of 0.80.Developed oversampled and undersampled models using XGBoost, integrating smote and hyper, each achieving an F1 score of 96%.Engineered and implemented an ensemble stacked model with TensorFlow and XGBoost, incorporating deep learning techniques like convolutional neural networks and regularization, as well as ensemble methods such as gradient boosting and tree pruning, which enhanced predictive accuracy in fraud detection to achieve 83% precision.Developed and deployed a real-time fraud detection system utilizing XGBoost and Apache Kafka to convert static transaction data into a live streaming pipeline, achieving a 97% F1 score for accurate and reliable fraud prediction.
