The process combines Genetic Algorithms' evolutionary optimisation characteristics with the collective power of Random Forests to increase the effectiveness of fraud detection. The Genetic Algorithm defines the most significant features and best RF hyperparameters using a fitness function based on F1-score and AUC. Selected chromosome contain features of both approaches (subsets) and combinations of parameters Optimised inputs are used in a second stage to train the Random Forest classifier, progressively improving the accuracy of the model while reducing the rate of false positives. This hybrid approach enables adaptive feature selection, generalisation of models and improved interpretability, all essential aspects for complex imbalanced FinTech fraud datasets.