DS/CS MQP 2023 ~ 2024: A Data-driven Analytical Approach for Improving Endurance Runners’ Performance

Rudometkin, Ethan; Perham, Mason

Student Work

DS/CS MQP 2023 ~ 2024: A Data-driven Analytical Approach for Improving Endurance Runners’ Performance

Public Deposited

The paper introduces a comprehensive analytical framework for predicting runners' endurance performance, addressing limitations in existing research. It includes a broad range of data including running gait, physiological, and anthropometric features. The feature selection method employed combines filter, wrapper, and embedded methods. Specifically, the Maximum Relevance Minimum Redundancy (MRMR) filter method, Recursive Feature Elimination (RFE) with a Random Forest model, and Ridge Regression were used to optimize feature selection. This multi-faceted approach ensures the selection of the most relevant and non-redundant features. The paper evaluates a variety of models, including SVM, ANN, AdaBoost, XGBoost, and deep learning models, for predicting runners' performance. The approach of testing multiple machine and deep learning models was intended to overcome the limitations of prior research that often restricted their analysis to a few models. By employing a more comprehensive range of models, this research aimed to identify the most accurate predictive model for runners' endurance performance. The evaluation metrics suggest that the XGBoost model yielded the most favorable results based on the error metrics provided: Mean Squared Error (MSE): 2.07, Root Mean Squared Error (RMSE): 1.44, and Mean Absolute Error (MAE): 1.12. The XGBoost model achieved the lowest values across these metrics, indicating its superior performance in predicting the outcome variable with the least amount of error.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

Subject