The effectiveness of machine learning in classifying prostate cancer from high–dimensional microarray data is often constrained by feature selection instability and scalability challenges. To address these issues, we propose the Ensemble-based Distributed Filtering and Classification Model (EnD-FCM), which integrates stability-driven ensemble learning with distributed feature partitioning. EnD-FCM applies multiple complementary filters: Chi-squared, ReliefF, mRMR, INTERACT, Information Gain, Consistency–across distributed partitions, and quantifies stability using Fisher’s discriminant ratio and the Kuncheva index, then aggregates stable subsets through majority voting prior to ensemble classification. This approach balances computational efficiency and predictive accuracy, in contrast to existing methods that emphasize only one. Using Naïve Bayes, KNN, Logistic Regression, and MLP classifiers, EnD-FCM achieved 99.89% cross-validation accuracy on the GSE55945 prostate cancer dataset (54,675 probes, 126 samples). ReliefF offered the best runtime efficiency, while consistency-based filtering enhanced stability and reduced redundant features. These results demonstrate EnD-FCM’s ability to reduce dimensionality, preserve scalability, and improve predictive reliability, establishing it as a robust tool for biomarker discovery in cancer diagnostics.Ensemble-based Distributed Filtering and Classification Model...