Research Experience

1. Study on Arrhythmia Classification Using Ensemble Learning and Birds of Prey-Based Optimization

• Analyzed the MIT-BIH arrhythmia dataset and identified severe class imbalance, with minority arrhythmia types  significantly underrepresented like V-class and prone to misclassification;
• Extract category-specific feature subsets during preprocessing to enhance the model's ability to distinguish features.
• Designed an ensemble learning framework integrating Random Forest and AdaBoost to jointly balance bias and variance, improving overall robustness and accuracy;
• Introduced a Birds of Prey-Based Optimization strategy to optimize high-dimensional hyperparameters (e.g., tree depth, number of estimators, feature sampling ratios) under multiple performance objectives, including accuracy, precision, recall, and F1-score;
• Achieved an overall classification accuracy exceeding 98%, with recall for the minority V-class surpassing 97%, demonstrating effectiveness in handling imbalanced clinical datasets. 

• Conducted a comparative study of time-series forecasting models, including ARIMA, LSTM, GRU, CNN-LSTM, and Graph Convolutional Networks (GCNs), using daily trading data (2018–2023) from the CSI 300 constituent stocks; 
• Designed a dynamic GCN model incorporating residual fusion of graph and temporal features to enhance predictive accuracy; 
• Evaluated models on direction accuracy, RMSE, and Sharpe ratio; GCN outperformed benchmarks during volatile market periods (75.8% accuracy);
• Analyzed sectoral correlation and attention weight distributions to improve interpretability; 
• Results accepted for presentation at the International Conference on Intelligent Data and Information Processing