Developed a sophisticated network analysis system for S&P 500 stocks (2018-2022), employing advanced graph theory and machine learning techniques to visualize and analyze market interconnectivity patterns. The project revealed crucial insights into stock market behavior during the COVID-19 pandemic, identifying influential stocks and sector-based communities through correlation analysis and network metrics. Notable findings included the detection of scale-free properties in market networks and the quantification of sector-based trading patterns using Jaccard similarity coefficients.
Developed and implemented multiple reinforcement learning agents (Q-learning and SARSA) for the Text Flappy Bird environment, conducting comprehensive performance analysis and parameter optimization. The project achieved superior game performance compared to random baseline agents, with Q-learning demonstrating optimal results through systematic hyperparameter tuning and comparative analysis of learning algorithms.
Developed a comprehensive machine learning system for predicting Airbnb prices in New York City, achieving 62.87% accuracy on test data using XGBoost. The project incorporated extensive exploratory data analysis, feature engineering, and model optimization, comparing various regression algorithms including classical ML models and ensemble methods.
Developed an advanced ensemble learning system combining Transformers and Convolutional Neural Networks for masked face recognition, achieving 94.22% accuracy through innovative stacking techniques. The project enhanced existing methodologies by implementing optimized weight distribution and stacked generalization, demonstrating significant improvements over state-of-the-art approaches. Notable achievements included a 2.2% accuracy gain over traditional ensemble methods and robust performance across multiple datasets.
Developed a sophisticated link prediction system for actor co-occurrence networks using advanced machine learning techniques, achieving 76.42% prediction accuracy with logistic regression. The project successfully combined graph structural features with node attributes to predict missing edges in the network, demonstrating superior performance across multiple evaluation metrics including F1 score (0.7607) and AUC score.
Developed a sophisticated multi-class sentiment analysis system using BERT, achieving 80.69% accuracy on aspect-based opinion classification. The project successfully addressed class imbalance challenges through weighted loss functions and optimised model architecture, demonstrating robust performance in categorising opinions as positive, negative, or neutral.
Developed a semantic image segmentation system for post-hurricane disaster assessment using UAV imagery, achieving 62.65% accuracy with SeResNet architecture. The project successfully identified and localized 27 different asset classes in residential areas, providing crucial support for disaster relief efforts in the Houston region after Hurricane Harvey.
Developed a sophisticated predictive analytics system for charity fundraising campaign “SJ22”, analysing donation patterns of 79,469 donors to optimise solicitation strategies. The project utilised advanced statistical modelling to predict donation likelihood and amounts, enabling data-driven decisions for maximising campaign profitability through targeted solicitation.
Developed a comprehensive COVID-19 information dashboard using Microsoft Power BI, featuring interactive visualisations of testing laboratories, case trends, and predictive forecasts. The system incorporated dynamic filtering capabilities and real-time data integration, providing healthcare professionals with crucial insights during the pandemic.
Developed an innovative web-based diabetes detection system using CNN and transfer learning, achieving high accuracy on the Pima Indians dataset. The project successfully implemented novel techniques for transforming tabular medical data into image representations, demonstrating effective application of deep learning in healthcare diagnostics.