Unified Framework for Outlier Detection Using Ensemble-Based Learning Techniques

In this study, we propose a unified approach to anomaly detection that integrates density-based fuzzy clustering with ensemble learning to improve detection accuracy and efficiency. The framework utilizes multi-layer perceptron (MLP), support vector machine (SVM), and k-nearest neighbors (KNN) classifiers, leveraging their complementary strengths for robust anomaly detection. To ensure comprehensive evaluation, the model is validated on multiple datasets, including NSL-KDD and CICIDS 2017, demonstrating its generalizability across diverse environments. Optimization strategies, such as bootstrap aggregation, are incorporated to address computational overhead while preserving performance. Furthermore, metrics like Jaccard, Dice, and Kappa are employed to provide detailed insights into model performance, ensuring interpretative clarity. By integrating federated learning for privacy preservation and explainable Al techniques for transparency, this work addresses critical challenges in anomaly detection. The results obtained as Sensitivity 92.48%, Specificity 99.73%, Accuracy 99.53%, Precision 97.96% and Recall 92.48% highlight the model’s ability to deliver high accuracy, efficient computation, and equitable performance, making it a significant contribution to the field.