Profiling and Classification of IoT Devices for Smart Home Environments

The goal of this study is to create a strong categorization system specifically designed for Internet of Things (IoT) device profiling. The main goal is to supplement current studies that use a wide range of machine learning techniques to identify anomalous behavior in Smart Home IoT devices with an exceptionally high accuracy rate. The intended framework is positioned to play a crucial function in bolstering IoT security in the future because it is made to include several types of abnormal activity detection. Our technological motivation stems from IoT smart sensors’ high processing power and advanced connectivity capabilities. Notably, these sensors have the potential to be manipulated for malicious purposes only on a single sensed data point rather than the complete collection of collected data from sensors, such as temperature, humidity, light, and voltage measurements. Such a threat lowers the detection effectiveness of many machine learning algorithms and has a substantial impact on the accuracy of aberrant behavior detection. To identify occurrences of alteration in one specific data point among the four potential data points collected by a single sensor, we compared and used different classifiers in our investigations, including the Decision Tree Classifier, KNeighbors Classifier, Support Vector Classifier (SVC), Logistic Regression, AdaBoost Classifier, Random Forest with Extreme Gradient Boost (XGBRF) Classifier, Random Forest Classifier, Light Gradient Boosting Machine (LGBM) Classifier, Gradient Boosting Classifier, and XGB Classifier. The results showed that the Gradient Boosting Classifier algorithm using random search attained an 85.96% detection accuracy, indicating a somewhat lower vulnerability to such changes. As a result, the Gradient Boosting Classifier algorithm with random search was the foundation for the carefully constructed suggested framework, which used four hyperparameter tuning mechanisms for comparison.