Predicting Performance Analysis of System Configurations to Contrast Feature Selection Methods

Computer system applications can be highly customized by providing users the option to select features for a desired configuration. However, these changes in an application affect the performance of the system. Hence, it is quintessential to be able to understand which feature(s) will affect the performance the most. This would enable us to derive an efficient and desired configuration. It is also of utmost that System design testing be done to ensure an errorless development, for smooth deployment of these configuration features. In this process, being able to predict the performance of the system based on the selected features becomes important. In this paper we try to predict the performance of an SQLite application for relational database systems using a dataset and also try to find the significant features. We have used two popular methods in data science to achieve this end namely regression analysis and decision trees. In addition to predicting the performance we have utilised feature selection algorithms such as ANOVA, Principal Component Analysis (PCA), t-SNE and Recursive feature extraction in order to optimise our prediction and contrast these algorithms.