· Utilized the RDKit library to preprocess the dataset by extracting molecular fingerprints (Morgan, MACCS, EState) and 208 RDKit descriptors.· Applied Machine Learning models (Gradient Boosting Decision Tree, Random Forest Regressor, Neural Networks) to predict HOMO energy levels and Kovats retention indices, optimized parameters with grid search and cross validation.· Evaluated model performance with MSE, MAE, and R-squared and achieved approximately 10% reduction in MSE… Show more · Utilized the RDKit library to preprocess the dataset by extracting molecular fingerprints (Morgan, MACCS, EState) and 208 RDKit descriptors.· Applied Machine Learning models (Gradient Boosting Decision Tree, Random Forest Regressor, Neural Networks) to predict HOMO energy levels and Kovats retention indices, optimized parameters with grid search and cross validation.· Evaluated model performance with MSE, MAE, and R-squared and achieved approximately 10% reduction in MSE, enhanced prediction accuracy and assisted more precise decision-making in chemical analysis. Show less

· Achieved K-means, Hierarchical, and DBSCAN clustering algorithms to do customer segmentation and applied Mini Batch K-means to improve the efficiency of data processing by approximately 15%.· Improved the traditional SMOTE method for imbalanced datasets using Gaussian kernel function and Kernel Density Estimation (KDE) to enhance the representativeness and distribution accuracy of minority class samples, applied on multiple datasets and increased a nearly 10% improvement in macro F1… Show more · Achieved K-means, Hierarchical, and DBSCAN clustering algorithms to do customer segmentation and applied Mini Batch K-means to improve the efficiency of data processing by approximately 15%.· Improved the traditional SMOTE method for imbalanced datasets using Gaussian kernel function and Kernel Density Estimation (KDE) to enhance the representativeness and distribution accuracy of minority class samples, applied on multiple datasets and increased a nearly 10% improvement in macro F1 and F1 scores. Show less

· Collected and analyzed over 7000 market data points, utilized R or Python to do data processing, visualization and actionable insights to support investment advisors with precise market trend analysis.· Developed a tailored investment portfolio for a corporate executive in a public fund project by analyzing customer profiles and 3 years of fund data, achieved over 12% ROI for the client in six months.