Comparison of Machine Learning Algorithms for Detecting Software Aging in SQL Server
Abstract
Software aging is a phenomenon characterized by the progressive degradation of system performance, resulting from the accumulation of internal erros, such as memory leaks and resource exhaustion. Efficient detection of this process is essential to prevent critical failures in production environments. Although several studies use Machine Learning (ML) algorithms to detect software aging, systematic comparison between these algorithms is still limited, especially in terms of their ability to predict resource exhaustion. This paper aims to fill this gap by comparing ML algorithms for detecting software aging, focusing on RAM memory exhaustion as the main indicator. The analysis was conducted using a dataset on RAM memory usage in SQL Server, applying the algorithms K-Nearest Neighbors (KNN), Support Vector Machine (SVM), and Random Forest (RF). The performance of the models was evaluated using the metrics Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and coefficient of determination (R2). Based on these indicators, it was possible to identify the most accurate algorithm and predict the time until memory exhaustion.
