Detecting Outliers with Semi-Supervised Machine Learning: A Fraud Prediction Application

Abnormal pattern prediction has received a great deal of attention from both academia and industry, with applications that range from fraud, terrorism and intrusion detection to sensor events, medical diagnoses, weather patterns, etc. In practice, most abnormal pattern prediction problems are characterized by the presence of a small number of labeled data and a huge number of unlabeled data. While this points most obviously to the adoption of a semi-supervised approach, most empirical studies have opted for a simplification and treated it as a supervised problem, resulting in a severe bias of false negatives. In this paper, we propose an innovative methodology based on semi-supervised techniques and introduce a new metric the Cluster-Score for abnormal homogeneity measurement. Specifically, the methodology involves transmuting unsupervised models to supervised models using the Cluster-Score metric, which defines the objective boundaries between clusters and evaluates the homogeneity of the abnormalities in the cluster construction. We apply this methodology to a problem of fraud detection among property insurance claims. The objectives are to increase the number of fraudulent claims detected and to reduce the proportion of claims investigated that are, in fact, non-fraudulent. The results from applying our methodology considerably improved these objectives.