It is obvious that football is the most popular sport in the world today, played in most countries in the world. The game provides a very large betting industry. The current estimations, which include both the illegal markets and the legal markets, suggest the sports match-betting industry is a multi-billion-dollar industry. This means the ability to accurately predict the outcomes of games can be very lucrative. Although various researches have employed statistical and mathematical models to predict match results, none of the models can predict the exact number of goals to be scored by each team in a football match at any given time. In this research, an Artificial Neural Network (ANN) model was developed following the Machine Learning (ML) Life Cycle. The dataset was obtained from Football-Data.co.uk, a reputable football data website. The backward Elimination technique was used to select the dataset features for the proposed neural network model. The features were extracted and the dataset was then split into training and testing sets at 75% and 25% respectively. The neural network developed is a multi-layer perceptron classifier implemented by the MLPClassifier class in sklearn. The model was compiled with different parameters to find the model with the highest accuracy relative to the Mean Squared Error (MSE). The graph for the accuracy score and mean squared error was plotted and it showed the mean squared error was relatively the same for all the models. The model with the highest accuracy score was selected. The selected model has three (3) hidden layers consisting of 10, 10, and 10 neurons with Sigmoid Optimizer and tanh activation function. The model ran 1000 epochs and got an accuracy score of 97.92% with an MSE of 2.8644, implying that real-life games with unknown results can indeed be predicted with a high level of accuracy using machine learning.
Buy article PDF