An ensemble hyper-tuned model for IoT sensors attacks and anomaly detection

Abstract

The extensive promotion of Internet of Thing (IoT), provides assorted opportunities and benefits in wide aspect of our life but unfortunately, the IoT is associated with various kinds of vulnerability attacks and anomaly exploits. Security experts indicate voluminous risks enforced by the IoT devices in different aspects. So, the attacks and anomaly detection become a growing concern in this sector. Disparate attacks allied Malicious Control, Denial of Service, Malicious Operations, Scan, Data Type Probing, Wrong Setup and spying becomes the severe cause of IoT system failure. The main objective of these attacks has to steal the confidential information from the system and generates unavailability of the system for authorized users. As compulsion of IoT security, we proposed a novel ensemble hyper-tuned model that automatically and effectively detects IoT sensors attacks and anomalies. This robust model is built on the basis of feature selection and ensemble technique of Machine Learning. The virtual IoT sensor’s environment generates a dataset by Distributed Smart Space Orchestration System (DS2OS), that is used for performing the experiments of attack detection by hyper-tuned Gradient Boosting ensemble algorithm. First, the feature selection process is applied for reducing the dimensions of the dataset, which enrich the environment for attacks and anomaly detection. After that, Gradient Boosting ensemble algorithm is applied with little hyperparameter tuning for getting the best results. Our model is outperformed for detecting the attacks on IoT sensor’s environment and effectiveness of this model is measured in terms of Accuracy = 99.40%, Precision = 99%, Recall = 99%, F1-Score = 99%. whereas, the ROC-AUC curve and confusion matrix is also generated for predicting the efficiency of our ensemble model.