SVM classifier optimisation for predicting anomalies in login data with SMOTE

The global cyberattack increase underscores the importance of more robust cybersecurity measures, particularly in detecting account takeovers. These incidents occur when unauthorized parties access user accounts, potentially leading to financial losses, data breaches, and reputational damage. The primary challenge in detecting account takeovers lies in their similarity to legitimate network traffic, making them difficult to identify using traditional methods. This research proposes using the Support Vector Machine (SVM) algorithm with a linear kernel optimized through RandomizedSearchCV to detect anomalies in login data. The dataset includes attributes such as IP addresses, geographical locations, network providers, device types, login times, and indicators of suspicious activity. Preprocessing uses the Interquartile Range (IQR) to remove outliers and the Synthetic Minority Oversampling Technique (SMOTE) to balance minority class data. This process ensures optimal data quality before further analysis. The research findings demonstrate that the SVM model achieves precision, recall, F1 score, and accuracy of 96%, proving its effectiveness in detecting anomalies and handling imbalanced data.