| Predictive data mining can be defined as the process of analyzing data and developing models for prediction. Predictive models and their ensemble strive to achieve highest accuracy possible in least training time. Combining models or classifiers ensemble, though improves the classification accuracy, use voting or variations thereof to reconcile models prediction. This, requiring the discretization of continuous value domain, decreases the prediction accuracy by generalizing the output domain to class interval mean. Besides this, training time of classifiers ensemble increases linearly with the number of models in the ensemble. This work proposes an ensemble model for reducing the generalization error and training time of prediction models. In this model the processes are cascaded sequentially eliminating the need of reconciliation of multiple predictions and thus of output domain discretization. The results of investigation on different problem domains demonstrate the predictability efficacy (higher accuracy and lower training time) of the proposed cascaded model. |
