Contributed to the development of AI-powered medical imaging solutions by performing end-to-end data analysis tasks. Cleaned, organized, and annotated large medical imaging datasets (X-rays, CT scans) to support machine learning model training. Conducted statistical analyses to assess data quality and identify trends, and collaborated closely with product managers and machine learning engineers to translate data insights into actionable improvements. Maintained detailed documentation to ensure alignment with regulatory and quality management system (QMS) standards.

This project explores the statistical insights offered by FIFA 24 player data, focusing on two main problems: predicting player value and country based on in-game characteristics, and comparing in-game valuations with real-world transfer fees. The dataset, sourced from Kaggle and supplemented with real transfer data from European leagues, was cleaned and preprocessed to ensure reliability. Missing values were addressed, irrelevant variables removed, and features such as composure and reactions identified as highly correlated with player value.

For Problem 1, machine learning models including Gaussian Naive Bayes, Decision Trees, Random Forest, Gradient Boosting, K-Nearest Neighbors (KNN), and Polynomial Regression were applied. Results showed that predicting a player's country from in-game stats was largely unsuccessful (accuracy ~10%), suggesting little correlation between nationality and abilities. However, predicting player value was more successful, with polynomial regression (degree 4) achieving ~78% accuracy.

For Problem 2, real transfer fees were compared with in-game values using statistical tests. The Mann-Whitney U-test showed significant differences, indicating that FIFA values do not directly match real-world fees. Linear regression revealed only a slight positive correlation, meaning higher in-game values loosely align with higher real transfer fees but lack precision.

Overall, the study highlights the usefulness of FIFA data for value prediction, while also revealing its limitations in replicating real-world economics.

his project investigates which in-game statistics most strongly influence winning in the NBA. Using Kaggle's “NBA Database” (game.csv), the team focused on regular season games and selected key variables—field goals attempted, free throws attempted, rebounds, steals, turnovers, and personal fouls. To standardize comparisons, they calculated stat differences between home and away teams and coded outcomes as win/loss (1/0).

The analysis began with T-tests, which confirmed all selected variables were statistically significant in distinguishing wins from losses. Several machine learning classifiers were then trained and tuned, including Gaussian Naive Bayes, Decision Trees, Random Forests, K-Nearest Neighbors (KNN), and Gradient Boosting. While Decision Trees and Random Forests risked overfitting, KNN and Gradient Boosting achieved the strongest validation accuracy, around 81%.

Feature importance analysis consistently identified rebounds as the most influential factor in predicting game outcomes, while turnovers also played a meaningful role. Conversely, personal fouls and free throw attempts were found to have minimal impact. Linear regression further supported these findings, showing a positive relationship between rebounds and win rate, and little correlation for personal fouls.

The study concludes that rebounds are the most significant stat in winning NBA games, offering valuable insight into team strategy, despite limitations such as outdated data and inclusion of former teams.