Modern-day football is becoming more and more reliant on an analytical approach, particularly in the optimization of scoring capabilities. The increasing volume of data from football events have set up numerous new research paths, and that can assist in the formulation of more informed decisions during match. Expected goals (xG) models can be utilized in different ways to assess performance and designing optimal strategies. This research aims to present a comparative and spatial analysis of some xG models to forecast football shot outcomes. In this context, several important features are employed to train different machine learning models including XGBoost, random forest, k-nearest neighbor (kNN), gradient boosting machines (GBM), multinomial logistic regression and Naïve Bayes (NB) focusing on predicting football shot outcomes to quantify the likelihood of a shot being a goal by the expected goal. The results have shown RF, XGBoost, and kNN performed better in terms of balanced accuracy and Kappa’s benchmark scale. In other words, the results of this analysis can be effective in competitive sports training including player assessment, team analysis and progress tracking, as it could assist football players to enhance their performance.