YUNRU ZHU

• Led the development of sales data dashboard software, resulting in a 30% increase in data processing speed and contributing to a 25% improvement in user satisfaction through three iterations of the dashboard.
• Optimized data updates and cleaning processes, reducing data processing time by 40% and minimizing errors, leading to a 15% increase in overall system efficiency.
• Enhanced the dashboard's user interface, which improved the user experience score by 20%, contributing to a higher adoption rate among users.

Project: Investment Portfolio Optimization

• Optimized an investment portfolio model using 20 years of U.S. stock data to assist Australian portfolio managers.
• Conducted feature selection through custom-designed equations.
• Tuned LSTM models for stock performance prediction and weight calculation.
• Applied gradient descent to optimize stock purchase weights.

Project: Modeling and Predicting Virtual Internship Evaluation Scores

• Conducted data cleaning and analysis to support the development of a predictive model for virtual internship scores.
• Classified and cleaned extensive textual data using techniques like Word2Vec and TF-IDF.
• Utilized visualization methods, including box plots and scatter plots, to analyze and present data insights.

Project: Predicting Short-Term Electricity Consumption

• Led data modeling to predict electricity consumption based on 20 years of weather data.
• Applied various regression models, including linear regression, random forest, and logistic regression.
• Achieved an R² exceeding 0.5, helping to mitigate power shortages and surpluses.

Project: Predicting ICU Patient Mortality Rates

• Led data modeling efforts to predict mortality rates for ICU patients using twenty years of vital signs data.
• Utilized multiple linear regression to identify key relationships between patient variables and mortality rates.
• Applied Principal Component Analysis (PCA) to reduce dimensionality and improve model efficiency.
• Conducted cross-validation to ensure model generalisability and prevent overfitting.
• Employed regularization techniques to enhance model accuracy and robustness.
• Achieved an R² of 0.65+, demonstrating strong predictive performance and providing actionable insights into patient outcomes.