Unsupervised Machine Learning Model for The Real-Time Prevention of Phishing

Every internet user is vulnerable to phishing or other forms of cyberattacks that, if care is not taken, using technology to store and secure soft assets would lose its essence since cybercriminals are devising new means of stealing these assets with minimum efforts. As of today, the world is struggling to overcome the menace of cybercrime. Phishing continues to be a crippling problem for many of today’s Internet users as the attack continuously escalate in figures and sophistications. This study developed an unsupervised machine learning model that prevents real-time phishing in the financial sector. The methodology adopted in this study is a hybrid unsupervised ensemble methodology which combined Autoencoders, Generative Adversarial Networks, and Self-Organizing Maps to model benign URL pattern, detect and block/prevent phishing attempts as statistical anomalies. The dataset needed for this study were gathered from Openphish and UCI. Model from Openphish dataset had accuracy of 94.48%, it performed below the model from UCI dataset with accuracy of 97.24%. Ensemble model of ANN and DL is highly efficient in detecting and preventing phishing urls in real-time. More specifically, ensemble model of deep learning and artificial neural network is the future of anomaly detection that researchers can even use in novel approach. Hence it was suggested that future studies should apply the ensemble model used in this study for spamming detection and deepfake detection and prevention.