Development of a Machine Learning-Based Model for Phishing Attack Detection in Smart-Home Environments

Internet of Things (IoT)-powered smart home solutions are increasingly prevalent. As a result, they are significantly altering how we live today. Due to device heterogeneity, resource constraints, user awareness, and other factors, the systems that connect the Internet of Things (IoT) are themselves connected to other devices, making phishing attacks one of the most significant hazards in smart homes. Phishing undermines the privacy, security, and integrity of smart ecosystems by exploiting a user's trust through phony emails, voice squatting, URLs, among others. Since the current security techniques are insufficient, this study aim is to create a new tool based on machine learning (ML) and deep learning (DL) techniques to check and detect phishing attacks on smart home systems. This study makes use of enron dataset from Kaggle which contains a total of 33,715 email samples. Data pre-processing was carried out to extract relevant features and attributes from the dataset. The phishector.py script provides a command line interface for processing emails and obtaining predictions. The model was evaluated using accuracy, precision, recall, and FI-score as performance metrics. From the obtained results, bagged decision tree model shows a 99.94% accuracy, precision 99.8%, F1-score, 99.94%, and recall, 100% while Adobos a 97.98% accuracy, precision 96.23%, F1-score, 98.08%, and recall, 100%. The study's result has further demonstrated how the suggested methods can be applied to detect phishing assaults in smart homes and shield linked smart devices from cyberattacks.