Resumen de Advanced Oversampling for Improved Detection of Software Anomalies in a Robot
Nuño Basurto Hornillos, Michał Woźniak 
, Carlos Cambra Baseca , Álvaro Herrero Cosío
Anomaly detection has been a challenging topic for decades and it still is open to new contributions nowadays. More specifically, the detection of anomalies (not only hardware ones but also those affecting the software) suffers from many problems when monitoring cyberphysical systems. One such usual problem is the much fewer data samples of anomalies than those available for the normal functioning of systems. This class-imbalance problem is addressed in the present paper and a novel strategy for oversampling the minority class is applied to an open dataset containing information about the performance of a component-based robot. The proposed strategy mainly consists on selecting the instances to be oversampled according to different criteria instead of randomly oversampling. Obtained results demonstrate that the proposed strategy improves predictive performance, especially when the SVM (Support Vector Machine) is used as classifier.
