Clustering to Determine Predictive Model for News Reports Analysis and Econometric Modeling

A tree model is constructed for the econometric problem domain and for topic modeling of news reports using a clustering approach. Here segments are represented as discretized intervals defined on econometric variables for speeding up the construction of regression tree. This discretization is achieved from variances defined on variables with predictability for that generated for calculating category utility values defined on correlated variables where the discretization method proposed has the aim to satisfy a constraint of minimum entropy distribution of values of the predictor variable among the categories. An algorithm is proposed for tree merging which is used for incrementally incorporating information for new time intervals with the existing model to generate updated tree model for maintaining logical consistency. The tree merging algorithm has been shown to be suitable for applying to news report documents or econometric information. This is accomplished with a proposed Pruning procedure for maintaining logical consistency in the merged tree which is applied together with existing approaches for limiting pruning and access costs for reducing misclassification error.