Determination of Insurance Policy Using Neural Networks and Simplified Models with Factor Analysis Technique

In this paper, we use feed forward neural networks with the back-propagation algorithm to build decision models for five insurances including life, annuity, health, accident, and investment-oriented insurances. Six features (variables) were selected for the inputs of the neural networks including age, sex, annual income, educational level, occupation, and risk preference. Three hundred insurants from an insurance company in Taiwan were used as examples for establishing the decision models. Six experiments were conducted in this study. These experiments were mainly categorized into two phases: Phase 1 (Experiments 1 to 3) and Phase 2 (Experiments 4 to 6). In Phase 1, we used the six features as the inputs of the neural networks. In Phase 2, we employed the factor analysis method to select three more important features from the six features. In Phase 1, Experiment 1 used a single neural network to classify the five insurances simultaneously while Experimental 2 utilized five neural networks to classify them independently. Experiments 1 and 2 adopted the purchase records of primary and additional insurances as experimental data. Experiment 3, however, utilized the primary insurance purchase dada only. In Phase 2, we repeated the similar experimental procedure as Phase 1. We also applied statistical methods to test the differences of the classification results between Phases 1 and 2. Discussion and concluding remarks are finally provided at the end of this paper. Key-words: Insurance policy, Neural networks, Back-propagation algorithm, Classification, Factor analysis, Feature extraction.