Feature selection for imbalanced data with deep sparse autoencoders ensemble

Class imbalance is a common issue in many domain applications of learning algorithms. Oftentimes, the same domains it much more relevant to correctly classify and profile minority class observations. This need can be addressed by feature selection (FS), that offers several further advantages, such as decreasing computational costs, aiding inference interpretability. However, traditional FS techniques may become suboptimal presence strongly imbalanced data. To achieve advantages this setting, we propose filtering algorithm ranking importance on basis reconstruction error deep sparse autoencoders ensemble (DSAEE). We use each DSAE trained only majority reconstruct both classes. From analysis aggregated error, determine features where presents different distribution values w.r.t. overrepresented one, thus identifying most discriminate between two. empirically demonstrate efficacy our experiments, simulated high-dimensional datasets varying sample size, showcasing its capability select generalizable class, outperforming other benchmark methods. also briefly present real application radiogenomics, methodology was applied successfully.