Football Player Performance Analysis using Particle Swarm Optimization and Player Value Calculation using Regression

Abstract

This paper aims at analysing the statistical data of various football players to establish a correlation between their play style and individual scores with their quantifiable attributes. Having established a correlation, the next step involves using the Particle Swarm Optimization(PSO) to simulate a match and draw a comparison between two random players, constraining their attribute scores within the boundaries of the top-recommended player for each attribute, as suggested by the k-nearest neighbors algorithm. This aids in setting up a benchmark score for the particular player position for a random selection from this subset based on z-score inference. Having optimized the player position score, stepwise regression and smoothing splines are used to model a prediction to compute the overall score of the player. Lastly, a regression equation is modelled using stepwise regression to estimate the net worth of the player based on their skill set, and predictions are performed using the optimal score obtained from PSO, by extracting the individual attribute scores from the inverse regression relation. From the experiment, the optimized score for the left striker(LS) comes out to be 86.32766. Running the PSO on all left strikers gives a 98% probability of obtaining a player whose score is greater than the benchmark score. For the two left strikers whose scores were optimized, the predicted worth from the stepwise model comes out to be 8.933734 and 8.191562, the former being greater than the historical worth.