Abstract: Detection of insurance premium fraud is provided by a predictive model, which uses derived variables to assess the likelihood of fraud for each policy. The predictive model produces a score, which is a measure of the likelihood of premium fraud or abuse. The predictive model is included in a system that accepts policies to be considered for scoring, selects which policies are appropriate for scoring, stores data about the policies in a database, uses the data to derive variables for the model, and processes and outputs the model scores and related information. A rule-based analysis, which detects specific inconsistencies in the data that are indicative of premium fraud, may also be part of the system. The model scores and red-flag indicators from the rule-based analysis may be further processed to provide customized output for users.