Enhancing IoT security: ensemble machine learning model for botnet attack detection
Annavaram Kiran Kumar, Padamati Pavithra, B. Yamini, Veluru Yashwitha, Thagupparthi Venkateswari · 2025
The largest number of interconnected ecosystems emerged from the quick spread of Internet of Things (IoT) devices, but this expansion has also left IoT environments susceptible to advanced cyberthreats, especially botnet attacks. These attacks cause operational problems and pose serious security and privacy issues by taking advantage of flaws in IoT networks. As botnet attacks become more complex and wide-ranging, conventional detection techniques are unable to detect them effectively and precisely in real time. Hybrid machine learning was created to overcome these difficulties, using the UNSW_NB15 dataset as a standard. There were several preprocessing techniques used to improve feature selection and model performance, including standardization, feature encoding, column transformer, one hot encoder and standard scaler. Random Forest obtained an accuracy rate of 95%, extra trees 94.85%, Decision Tree 93.69%, multilayer perceptron (MLP) 93.44%, Gradient Boosting 93.15%, K-nearest neighbors algorithm 92.91%, and logistic regression 91.07%. Long short-term memory (LSTM) was the most effective algorithm for detecting seasonal patterns in network traffic with an accuracy of 96.435% in training and 96.665% in testing. It provides real-time botnet attack alerts by allowing users to submit CSV data files with network traffic using Streamlit.