The inherent uncertainty in betting games makes it impossible to accurately predict results, especially in games of chance like poker. However, in sports, bets are not so random, and results can be predicted by analyzing large volumes of data using stream-based machine learning models. Specifically, we propose a sports betting system that analyzes the results of tennis matches. We performed a multivariate analysis to engineer features and generalize the model training process. We have also exploited a time window series approach to obtain accumulated values from several matches. Interpretability is also provided through counterfactual explanation techniques. Experimental results are close to 72% in accuracy. Ultimately, our solution allows for reducing uncertainty in sports analyses.