Sahil Sangani

Led an end-to-end data science project predicting MLB player WAR (Wins Above Replacement) using advanced machine learning techniques, winning 1st place in a competitive 8-week challenge.

• Won 1st place at UIUC Illinois Data Science Club's Data Dive competition, competing against 50 teams over 8 weeks
• Led complete project lifecycle from problem formulation and data collection to model evaluation and presentation
• Applied Principal Component Analysis (PCA) for dimensionality reduction, identifying key performance indicators from complex baseball statistics
• Implemented Gradient Boosting model optimized for both predictive accuracy and interpretability, enabling actionable insights for team management
• Presented results to panel of academic and industry judges, effectively communicating technical methodology and business value
• Emphasized model interpretability to provide clear explanations of which player attributes drive value predictions

Built predictive models for fourth-down conversion success using NFL play data, comparing neural networks and logistic regression to evaluate model complexity trade-offs.

• Developed comparable models achieving AUC ~0.68 using both neural networks and logistic regression
• Analyzed real-world NFL play-by-play data to identify factors influencing fourth-down conversion success
• Conducted comprehensive model comparison, finding limited performance gains from increased complexity on noisy real-world data
• Demonstrated practical understanding of model selection trade-offs: interpretability vs. complexity
• Applied insights relevant to NFL coaching decisions and game strategy optimization

A machine learning project focused on mental health risk assessment for students, combining statistical analysis with predictive modeling to identify at-risk individuals.

• Achieved exceptional model performance with AUC of 0.936, indicating strong discriminative ability between risk categories
• Performed comprehensive feature engineering on student survey data, transforming raw responses into meaningful predictors including sleep patterns, financial stress indicators, and academic pressure metrics
• Conducted in-depth coefficient analysis to identify the most significant risk factors, providing actionable insights for intervention strategies
• Collaborated effectively in a team of four, coordinating data preprocessing, model training, and validation tasks
• Designed model with real-world application potential for school counseling services and student support programs

An advanced sports analytics project that leverages Statcast data to classify baseball pitches in real-time, demonstrating the intersection of machine learning and sports technology.

• Developed highly accurate pitch classification model achieving 98.6% accuracy using K-Nearest Neighbors algorithm on Kevin Gausman's Statcast data
• Engineered sophisticated features including release speed, spin rate (rpm), horizontal and vertical movement, and release point coordinates to capture pitch characteristics
• Created comprehensive data visualizations using Seaborn to demonstrate pitch separability and model decision boundaries, enhancing interpretability for baseball analysts
• Built modular, scalable framework that can be adapted for any MLB pitcher by simply substituting the dataset, making it practical for broadcast integration
• Optimized hyperparameters through cross-validation to ensure model generalization across different game situations

A data-driven analysis exploring how lifestyle and demographic factors influence tumor size using regression and ANOVA modeling in R.

• Built multiple linear regression and two-way ANOVA models to analyze relationships between lifestyle factors and tumor size
• Achieved R² = 0.68, explaining 68% of tumor size variability through behavioral and demographic predictors
• Designed a dynamic, templatized R Markdown pipeline that automatically updates analyses and visualizations when new datasets or variables are used
• Created interactive visualizations with ggplot2 to highlight main and interaction effects among predictors
• Simulated lifestyle-change scenarios to assess predicted health improvements

A collection of exploratory data analysis and regression modeling projects across different domains, showcasing versatility in data science techniques and problem-solving approaches.

• Analyzed college tuition dataset to understand pricing patterns, exploring relationships between institution characteristics, location, and tuition costs
• Conducted used car price analysis, investigating how factors like mileage, age, brand, and features influence market value
• Created publication-quality visualizations using Matplotlib and Seaborn, effectively communicating insights
• Built interpretable linear regression models with strong predictive performance (R² = 0.83)
• Practiced comprehensive data preprocessing including handling missing values, outlier detection, and feature scaling