Financial Application of Multi-Instance Learning: Two Greek Case Studies

Abstract

The problems of bankruptcy prediction and fraud detection have been extensively considered in the financial literature. The objective of this work is twofold. Firstly, we investigate the efficiency of multi-instance learning in bankruptcy prediction. For this reason, a number of experiments have been conducted using representative learning algorithms, which were trained using a data set of 150 failed and solvent Greek firms in the recent period. It was found that multi-instance learning algorithms could enable experts to predict bankruptcies with satisfying accuracy. Secondly, we explore the effectiveness of multi-instance learning techniques in detecting firms that issue fraudulent financial statements (FFS). Therefore, a number of experiments have been conducted using representative learning algorithms, which were trained using a data set of 164 fraud and non-fraud Greek firms. The results show that MIBoost algorithm with Decision Stump as base learner had the best accuracy in comparison with other multi-instance learners and single supervised machine learning techniques.