Applying Machine Learning Techniques to Advance Anti-Doping

Globally exists an ongoing battle between increasingly advanced doping methods and limited resources available to anti-doping organizations. Therefore, the developments of new tools for identifying athletes who may be doping are needed. Recognizing which athletes are at the highest risk of doping allows an anti-doping organization to distribute those limited resources in the most effective manner. Presented below is a comparison of multiple machines and statistical learning approaches, combined with resampling techniques, to identify which athletes are at the highest risk of doping. The results presented indicate that support vector classification and logistic regression, combined with oversampling, may provide an effective tool to aid anti-doping organizations in most effectively distributing scarce resources. Adoption and implementation of these methods may both enhance the deterrence effect of anti-doping, as well as increases the likelihood of catching athletes doping. Future research should be conducted to explore the feasibility of combining these performance-based measures with biological measures such as the Athlete Biological Passport to enhance anti-doping efforts.