Validated Agent-Based Model using Predictive Data Mining and Intervention Policy Testing Framework: A Case Study in Child Vehicle Safety

Much work has been done on making and perfecting agent-based simulations on child safety measures in cars. These simulations, using algorithms based on social networks, cultural algorithms etc. try and predict what factors are responsible for the propagation of knowledge about child safety measures in a given society. One of the biggest factors being over-looked in these simulations is the validity of the model. In absence of validation against real data, these models may not be a true representation of a real world scenario, and the trends predicted though these simulations are questionable. This paper proposes a system design using regression analysis and predictive data mining on a survey done in the field of child safety. Using the result of this data mining process in the form of a decision tree, we can initialize our agent-based model with data from the survey and later validate the model comparing the results to the survey data. Consequently a framework is formed to test different agent profile based intervention techniques, so that a decision about selecting an intervention technique with a given cost can be demonstrated.