Personalized Training Optimization for Sports Athletes Using a Hybrid Machine Learning and Rule-Based Expert System Approach

Abstract

The current research provides a basis for the personalization of athletic training optimization, but there are still many possibilities for the further extension and refinement of the hybrid intelligent system approach for future research. The current system includes each athlete assessment as an athletic snapshot in isolation and does not have information about the temporal aspects of training history and response to training. In the future this type of work should also include modelling approaches such as recurrent neural network or state space models that observe the evolution of performance over longer training cycles. Longitudinal data would allow the system to adapt learned individual adaptation rates, recognize the individual as a responder to or non-responder of particular training stimuli and enact adaptive programming, the means to adjust based on the results actually achieved rather than static predictions. The use of closed-loop feedback control using the frameworks of reinforcement learning is a very promising direction. The ability to discover nearly optimized training adjustment policies by allowing the system to see outcomes in many thousands of combinations of athlete and intervention opens the door to finding some of the subtler interaction effects between aspects of the athlete and the program variables that are difficult to encode as rules. This approach would make the system from prescriptive to truly adaptive, so that recommendations would improve as more and more athletes move through training cycles.