Sandy Mulia Kesuma, Fajri Profesio Putra
Football formation analysis has become increasingly data-driven to assess team formation suitability and support the coaching process, since intuitive formation selection has proven inefficient and lacks objectivity. This study proposes a football formation recommendation system using the Gaussian Naïve Bayes algorithm, based on team attribute aggregation and expert tactical weighting. The dataset consists of 661 outfield football players' attributes, summarized into six core attributes: Pace, Shooting, Passing, Dribbling, Defending, and Physical. Using the Knowledge Discovery in Databases (KDD) methodology, these attributes were aggregated at the team level and tactically weighted based on interviews with a professional football coach, producing three ratio features (Attack, Midfield, Defence) as the basis for classification. The model was evaluated using 5-fold cross validation, confusion matrix analysis, and ROC curve analysis to assess its discriminative ability. Evaluation results show an average accuracy of 82.8%, precision of 91.07%, recall of 89.53%, an F1-score of 86.71%, and an AUC of 0.715. A web-based application was built using Laravel, integrated with a Python prediction module, allowing coaches to upload team data and obtain formation recommendations along with their success probabilities.