CycleFit: An Analysis of Regression Models for Caloric Expenditure Prediction in Cycling Activities

In the realm of endurance sports, athletes often grapple with the challenge of balancing caloric intake to avoid the dreaded energy depletion phenomenon known as "bonking." This study delves into the realm of personalized data analysis, utilizing cycling data sourced from the Strava platform Utilizing a dataset containing labeled energy outputs and various activity features, regression techniques are employed to model the relationship between these features and energy expenditure in kilojoules. The study explored multiple machine learning models, each evaluated based on its R-squared (R2) score. Notable models and their respective scores include Support Vector Regression with a linear kernel (R2 = 0.97667), CatBoost (R2 = 0.9950), Random Forest (R2 = 0.9879), Linear Regression (R2 = 0.9754), XGBoost (R2 = 0.9945), Decision Tree (R2 = 0.9746), and k-Nearest Neighbors (R2 = 0.8651). These high R2 scores highlight the effectiveness of the predictive tool, showcasing strong correlations between selected features and energy expenditure. The tool provides cyclists with a practical means of estimating caloric needs during rides, offering a personalized approach to nutrition and helping prevent the detrimental effects of energy deficit during endurance activities.