Abstract

The adoption of IoT has become important due to the effortless access and remote management of things in the environment. This is made possible through the integration of web-based systems into IoT, allowing accessibility through the internet but due to this, there is always a chance get being targeted by the attacker by finding the security vulnerabilities in the edge or the cloud-based integration. In this work, We present a model designed to classify anomalous activities within a web-based IoT system. Our approach encompasses binary classification, achieved through an efficient reduction of the feature space. through this, our system enhances interpretability and optimizes model efficiency, ensuring precise and effective classification. To obtain a condensed feature subset, our method combines Pearson correlation analysis with permutation-based feature importance. We assessed several machine learning models, including Logistic Regression(LR), K-Nearsest Neighbour(KNN), Random Forest(RF), and Xgboost(XGB), for binary classification tasks using the CICIoT2023 dataset. With just 10 selected features, Xgboost and Random Forest among these models frequently produced outstanding results, claiming 99% accuracy and a 98% F1 score. This demonstrates how successful and efficient the method is at categorizing unusual IoT Web-based activity.