Robust Methodology Design to Detect Anomalies Over Wireless Sensor Networks Using Predictive Learning Strategy
R Rajeshwari, Ajai Athithya, M Gowthamkumar, R Manigandamurthi, B Sivanesh, R Nivetha · 2024
In the realm of intrusion detection, the effective utilization of machine learning algorithms is paramount to safeguarding network security. Using the NSL-KDD dataset, this study investigates how well different feature extraction approaches and prediction models perform intrusion detection. We looked at a variety of feature extraction methods such ShuffleNet, CNN, PCA, and Autoencoder, as well as a variety of prediction models like LightGBM, Random Forest (RF), SVM, and Logistic Regression. The F1-score, recall, accuracy, and precision are some of the critical performance indicators used to evaluate each approach. The findings demonstrate that ShuffleNet is effective for feature extraction, especially when used in conjunction with LightGBM, which improves its testing accuracy to 0.92. Furthermore, the predictive models exhibit varying degrees of effectiveness, with LightGBM demonstrating superior performance compared to other models, achieving a testing accuracy of 0.92. Additionally, the hybrid model consisting of ShuffleNet and LightGBM achieves a remarkable testing accuracy of 0.94, highlighting the potential of integrating feature extraction and predictive modeling techniques to enhance intrusion detection systems and bolster network security against cyber threats.