AN OPTIMIZED WAY OF ANALYSIS AND PREDICTION OF INTRUSION DETECTION CICIDS2017 DATASET USING ML BASED STACKING CLASSIFIER

Abstract

The growing security threat of cyberattacks requires advanced intrusion detection systems capable of accurately identifying diverse threats with minimal false alarms. This study presents an optimized methodology for analysing and predicting network intrusions using the comprehensive CICIDS2017 dataset, which features realistic, contemporary network traffic. The current ML based algorithm includes CAT boost, Logistic Regression, Random Forest lacks Detection accuracy and efficiency parameters. The core detection mechanism deployed in this research employs a stacking ensemble classifier, a hybrid approach, utilizing high-performing, diverse models imbibing LightGBM+XGBoost+SVC. The implemented stacking classifier provides extensive results with Accuracy 99.9%, Precision 99.8%, Recall 99.7%, F1 Score 99.6 on provided input data which seems to be enumerate higher performance than the current available methodologies.