Predicting Health Insurance Policy Continuation Using the Naïve Bayes Classification Algorithm

The continuation of health insurance policies is relevant to premium stability and customer relationship management. This study presents an exploratory application of the Naïve Bayes algorithm to classify policy continuation at PT BNI Life using administrative information from 2019. The authors confirmed an archival dataset containing 65,785 policy records; however, the only recoverable model-evaluation evidence in the submitted materials concerns an earlier experiment with 666 observations and 133 test cases. Nine demographic, financial, and policy-related attributes were described, and the analysis followed the stages of the CRISP-DM framework. A corrected reading of the original RapidMiner confusion matrix gives 57.14% accuracy, 54.69% precision, and 55.56% recall for the continuation class. The original receiver operating characteristic figure reports an area under the curve of 0.57. An arithmetic audit also corrects the classification decision in the historical manual Naïve Bayes illustration. The results demonstrate that the algorithm was implemented in a limited experimental setting, but its predictive discrimination was weak. Because the selection of the evaluated observations from the archival records is not documented, the findings cannot be generalized to the complete portfolio. This case study offers a transparent account of the available evidence and identifies requirements for more reliable insurance classification research.