A Building Damage Classification Framework for Feature Subset Selection using Rough Set with Mutual Information

Authors

  • M. Vishnu Vardhana Rao, Aparna Chaparala

Abstract

Predictive analysis (PA) is one of the advanced analytics or decision systems for finding future predictions. It assesses the risk based on some conditions in a particular dataset, which used to make predictions about unknown future events. Prediction future outcomes and trends, PA used as model for extracting (inheritance) the information from existing datasets in order to determine patterns. Due to highly time complexity processing, researchers use standard datasets for predicting the unknown future outcomes and trends. However, the dataset consists of a set of features or sequences of attributes. The features in the dataset explain the total description of the datasets. Based on features in the dataset, the classification can occur, and some of the features not highly correlated with other features in the dataset. The inappropriate or avoidable or duplicating features tent to down the accuracy for solutions. From the above lines, the reduction of features or feature selection is a critical process for the classification job. The available features in the dataset selected to get better results in the classification process. The reduced attribute subset description is more suitable for classification. From now, attribute reduction or feature selection is an energetic method for classification responsibilities. This research proposes a new approach to reduce features or the attributes or the properties of the dataset based on Rough set (RS) with mutual information balance. This approach expected makes the reduction process efficient. Although, this approach is also not to prune to time complexity reduction. The use of the Rough Set (RS) theory predicts the importance of various features and certain critical features without additional information other than the necessary information. Hence, this work further refines the strategy to reduce the time complexity by deploying a wrapper feature selection approach. The results of this proposed framework are highly satisfactory and improve the classification results. The results of this algorithm tested on various standard classification methods, and the improvements are notable. The proposed method tested on various standard algorithms with the original and reduced feature sets, and it observed that the accuracy has increased with a reduction in time complexity.

Published

2020-02-29

Issue

Section

Articles