Predictive Modeling of Behavioral Outcome through Non Domain Expert Method Using Naive-Bayes Technique

Predictive models validate the anticipation of certain aspects of human behavior, such as goals, actions and preferences. In order to develop predictive models, there are many approaches in machine science which describes the computational techniques to develop models. This paper explains an approach that the non-domain experts can collectively design an attributes such that which subset of data to study and provide values for those attributes which helps to predict some behavioral outcome of interest. This is organized by allowing human groups to interact both by answering the questions and also pose questions their peers through web platform, likely to help in predictive modeling. However in literate linear regression technique is used to predictive modeling of behavioral outcomes. Regression creates predictive models with numerical or continuous target attributes. If the target attribute contains continuous values, it creates a regression problem that cannot explicitly refer to all target categories that are used in the model. The major challenge with this model is it over fit the data. This paper presents an improved method that uses Naive Bayes for modeling the prediction. Naïve Bayes overcomes the problems in earlier technique by dealing with discrete or categorical target attributes. Experimentally it is observed that the accuracy of Naive Bayes approach for behavioral prediction is better than the linear regression.