Prediction of dementia by increasing subspace size in rank forest

Abstract

Decision tree and decision forest are widely used as the best practices for knowledge discovery and prediction process. However, there is still room for improvement in terms of accuracy and processing time. The aim of this study is to improve the accuracy of knowledge discovery and uses extracted rule sets from knowledge discovery process to build a highly accurate prediction model with minimum processing time. We use a different combination of state of art algorithms like Random forest, rank forest and sub-spacing to discover interesting rules for knowledge discovery by using probabilistic values for splitting attributes in a tree and increase sub-space in a lower node to include all attributes that have potential information for knowledge discovery. Then use that knowledge to build a model for future prediction. Adding a different combination of decision trees and forest algorithms on the framework proposed by Zahid and Adnan resulted in the increased accuracy of knowledge discovery. Further research shows that use of probabilistic values for splitting attributes in a tree and increasing sub-space in a lower node of trees includes all attributes that have potential information for knowledge discovery. Thus, the proposed prediction model has decreased the processing time from 0.82 seconds to 0.11 seconds and increases the accuracy from 71.6% to 83.1%. In this paper, the different combination of rank forest and increased size of subspace is used in top of state of art. This approach has overcome the limitation of state of art and does not extract any partial rules and generate highly accurate rule sets with low processing time.