Feature Selection and Categorization to Design Reliable Fault Detection Systems

摘要: In this work, we will develop a fault detection system which is identified as classification task. The classes are the nominal or malfunctioning state. To decision it important to select among data collected by supervision system, only those carrying relevant information related There two objectives presented in paper, first one use mining techniques improve tasks. For purpose, feature selection algorithms applied before classifier measures needed for system. second objective STRASS (STrong Relevant Algorithm of Subset Selection), gives useful categorization: strong features, weak and/or redundant ones. This categorization permits design reliable algorithm tested on real benchmarks medical diagnosis and detection. Our results indicate that small number can accomplish perform task shown our ability detect correlated features. Furthermore, proposed efficient