Predicting football match outcomes: a multilayer perceptron neural network model based on technical statistics indicators of the FIFA world Cup

This paper utilizes the strong non-linear approximation capability of a multilayer perceptron Neural Network to predict match outcomes based on Technical Statistics Indicators. Principal component analysis was applied to all the official data for dimensionality reduction and feature identification, resulting 22 technical statistics indicators. An architecture of a Multilayer Perceptron Neural Network with a 24-4-3 was constructed using SPSS. The results showed that the model achieved an overall prediction accuracy of 86.7%, the prediction accuracy for Draw is substantially lower than for the Win and Loss. The neural network model exhibited robust predictive performance. On this basis, five relevant topics were discussed, including model performance evaluation, relationship between TSI and match outcomes, discriminative power of TSI, impact of stage on prediction results and incorrect predictions of match. Thus, coaches can enhance the team's performance-oriented results under limited training resources by transforming the high-impact technical statistical indicators identified by the model into training priorities, thereby achieving data-driven scientific training management.