Machine Learning for the Prediction of the Index of Effectiveness in Cycling

Biomechanical research permits the description, analysis, and assessment of the mechanics and biophysics of the musculoskeletal system during the execution of a movement. It is a science that is related to different fields such as medicine, sports, ergonomics, and engineering. It involves carefully analyzing each movement position, velocity, and acceleration to understand how the body moves and functions. Biomechanics has benefited from technological advances in recent years resulting in a large amount of data for kinematic and kinetic analysis. In this way, it has contributed by developing innovative, effective, and practical solutions to improve athlete performance, optimize training plans, improve movement patterns, prevent injuries, and develop better equipment. Nowadays, machine learning algorithms have transformed different aspects of the world, from modifying the process of performing tasks to health care and sports, increasing the performance of athletes without neglecting their health and comfort. Furthermore, machine learning is a solution to solve complex problems in biomechanics. Nevertheless, some of these techniques are often black boxes, in which the contribution of each predictor is unknown, which may be a disadvantage. In the context of cycling, there are many biomechanical studies applied to improve the cyclist’s performance and understand his movement technique. Pedaling technique is one of the most important and complex aspects of a cyclist because it affects his efficiency, power, endurance, and comfort. For this purpose, it is necessary to generate an effective force on the cranks through a combination of the cyclist’s muscle recruitment and joint kinematics. Currently, for measuring that, a few motion capture systems and instrumental devices have been developed. They make it possible to identify kinematics and kinetics variables, as well as power and cadence, to know if the pedaling force production is optimal. Considering these multiple variables are obtained from different measurement systems, it is difficult to analyze and compare analytically the results. Machine learning algorithms have been considered as alternatives for modeling in biomechanics, and there is no doubt that it has the potential to revolutionize the sports industry and push the boundaries of what is possible. The prediction of variables in cycling through machine learning allows a better understanding of the phenomena involved in the generation of pedaling force and power. Therefore, in this chapter, we present the methodology for the creation of an interpretable model for predicting pedaling efficiency from lower-limb kinematics. For this purpose, we have analyzed the strengths and limitations of some machine learning methodologies to propose the most optimal and interpretable model according to the knowledge of biomechanics. Therefore, the chapter is organized with the first part introducing the basic concepts of biomechanics. Then some machine learning concepts are presented with emphasis on multiple linear regression model. Subsequently, biomechanical modeling and some applications in cycling are presented. It also presents the methodology for the development of machine learning models. In addition, an example of the application of a methodology applied to cycling is illustrated. In the example, a multiple linear regression model for the prediction of the index of effectiveness is clearly explained with emphasis on the selection of predictors to obtain an interpretable model. Finally, an analysis of the model obtained with the conclusions of the example shown is provided, leaving open the possibility of applying other machine learning methodologies, not only to the case of cycling but also to other sports, thanks to the versatility of this type of methodologies.