Synapse Defender: A Hybrid Deep Learning and Machine Learning Approach for Intelligent Intrusion Detection

Authors: Khushal Patil, Sarthak Yere, Gaurav Pawar, Aryan Deore, Shilpa Khedkar

Country: India

Full-text Research PDF File:   View   |   Download

Abstract: Today’s network infrastructures are becoming more complicated due to the fast rise of cloud computing and IoT devices, which makes detecting cyber-attacks more difficult. The conventional intrusion detection systems (IDS), based on predefined rules or known attack patterns, tend to be very ineffective in detecting new or unknown attacks. In this paper, Synapse Defender is proposed, a meta-hybrid intrusion detector, is proposed in this paper, which involves using a CNN-LSTM-based feature extractor and stacking ensemble model with XGBoost, LightGBM, and Logistic Regression as a meta-learner. The model is evaluated on the CICIDS2017 dataset. The framework includes a three-stage feature selection process that reduces features from 78 to 50, RobustScaler preprocessing for handling outliers, and focal loss to address class imbalance without synthetic sampling. Moreover, per-class threshold tuning enhances the ability to detect minority types of attacks. On the whole, the proposed scheme improves the detection performance and helps to solve the major issues related to the imbalance of the classes, high-dimensional data, and efficient computing in the contemporary IDS.

Keywords: Intrusion Detection Systems (IDS), Deep Learning, CNN-LSTM, Ensemble Learning, XGBoost, LightGBM, Network Security, Anomaly Detection, CICIDS2017, Class Imbalance, Feature Selection, Cybersecurity

Paper Id: 233120

Published On: 2026-05-17

Published In: Volume 14, Issue 3, May-June 2026

Share this