Finding Originally Mislabels with MD-ELM

This paper presents a method which aims at detecting misla-beled samples, with a practical example in the field of bankruptcy predic-tion. Mislabeled samples are found in many classification problems and can bias the training of the desired classifier. This paper is proposing a new method based on Extreme Learning Machine (ELM) which allows for identification of the most probable mislabeled samples. Two datasets are used in order to validate and test the proposed methodology: a toy exam-ple (XOR problem) and a real dataset from corporate finance (bankruptcy prediction).