Using Machine Learning to Detect Unauthorized Access in Database's Log Files

The paper investigates the use of machine learning techniques to detect unauthorized access in database log files. Results show that most algorithms of supervised machine learning performed well in identifying normal cases but struggled to detect anomalies, with the exception of Naïve Bayes and Random Forest which gave mediocre results by identifying one out of twenty anomalies. In the semi-supervised machine learning methods, Local Outlier Factor showed an accuracy of 0.98 in detecting normal cases and 0.7 in detecting anomalies. One Class Support Vector Machine had an accuracy of 0.89 for normal cases and 0.05 for anomalies, while Isolation Forest had an accuracy of 0.98 for normal cases and 0.0 for anomalies. These findings suggest that semi-supervised techniques may be more effective in detecting unauthorized access in database log files.