A Comparative Study of Machine Learning Techniques for Predicting Sports Match Outcomes

Abstract

The increasing interest in predictive analytics for sports performance has encouraged the development of intelligent systems capable of forecasting match outcomes. This project presents a machine learning-based prediction system designed to determine the winning team based on historical match data. The system allows users to upload a dataset, visualize key trends such as successful versus unsuccessful chases, and perform essential data preprocessing including handling missing values, shuffling, normalization, and splitting into 80% training and 20% testing datasets. Multiple machine learning algorithms— including K-Nearest Neighbors (KNN), Random Forest, Artificial Neural Networks (ANN), and 2D Convolutional Neural Networks (CNN2D)— are implemented and evaluated. Each model is trained on the processed dataset, and its performance is assessed using accuracy, precision, recall, F1-score, confusion matrices, and ROC-AUC analysis. Among the algorithms, CNN2D achieved the highest accuracy of 96%, followed by Random Forest with 81% and KNN with 70%, while ANN showed moderate performance. A comparison module provides a visual and tabular summary of all models, enabling clear identification of the most effective approach. The system also supports real-time prediction on new test data, displaying the predicted winning team along with the expected score. Overall, this framework streamlines match outcome analysis, reduces manual evaluation, and provides a reliable tool for predictive decision-making in sports analytics.