kyle zhao

Statistical analysis of professional League of Legends match data to determine whether ADC players with gold advantages at 15 minutes significantly impact win probability. Collaborative UCSD DSC 80 project analyzing 10,000+ competitive matches from Oracle's Elixir dataset.

• ▸Analyzed 150,000 rows of professional esports data using hypothesis testing and permutation tests (p < 0.001)
• ▸Built Random Forest classification model achieving 75% accuracy with feature engineering and GridSearchCV tuning
• ▸Demonstrated teams with ADC gold leads achieve 65-70% win rates through statistical analysis
• ▸Validated model fairness across game scenarios using permutation testing (p = 0.156)

Exploratory and statistical analysis of the sitcom Friends to identify patterns in viewership, dialogue distribution, and emotional range to inform a potential reboot strategy.

• ▸Analyzed episode-level data including directors, writers, and viewership trends
• ▸Applied hypothesis testing and regression analysis to examine IMDB ratings and audience patterns
• ▸Used NLP techniques to analyze dialogue distribution and generate episode titles in the style of the original show

Built a logistic regression model to predict diabetes onset using medical record data, with a focus on data cleaning, visualization, and interpretability.

• ▸Cleaned and preprocessed medical data by handling missing and invalid values
• ▸Visualized feature distributions and correlations using histograms and scatter plots
• ▸Achieved 75% accuracy on a held-out test dataset using logistic regression

Collaborated on developing a convolutional neural network to classify images from the CIFAR-10 dataset, focusing on model optimization and performance evaluation.

• ▸Implemented and trained a CNN using TensorFlow
• ▸Experimented with hyperparameters to optimize validation performance
• ▸Achieved 79.74% test accuracy on CIFAR-10 image classification