A principal feature analysis

A key task of data science is to identify relevant features linked certain output variables that are supposed be modeled or predicted. To obtain a small but meaningful model, it important find stochastically independent capturing all the information necessary model predict sufficiently. Therefore, we introduce in this work framework detect linear and non-linear dependencies between different features. As will show, actually functions other do not represent further information. Consequently, reduction neglecting such conserves information, reduces noise thus improves quality model. Furthermore, smaller makes easier adopt given system. In addition, approach structures within considered This provides advantages for classical modeling starting from regression ranging differential equations machine learning. show generality applicability presented 2154 center measured classification faulty non-faulty states set up. number automatically reduced by 161 The prediction accuracy even compared trained on total second example analysis gene expression where 9513 genes 9 extracted whose levels two cell clusters macrophages can distinguished.