Multi-Level Dimensionality Reduction Methods Using Feature Selection and Feature Extraction

Abstract

This paper presents a novel feature selection method called Feature Quality (FQ) measure based on the quality measure of individual features. We also propose novel combinations of two level and multi level dimensionality reduction methods which are based on the feature selection like mutual correlation, FQ measure and feature extraction methods like PCA(Principal Component Analysis)/LPP(Locality Preserving Projection). These multi level dimensionality reduction methods integrate feature selection and feature extraction methods to improve the classification performance. In the proposed combined approach, in level 1 of dimensionality reduction, feature are selected based on mutual correlation and in level 2 features are selected from the reduced set obtained from level 1 based on the quality measure of the individual features and vice versa. In another proposed combined approach, feature extraction methods like PCA and LPP are applied on the reduced feature set obtained in level 1 or /and level 2. To evaluate the performance of the proposed methods several experiments are conducted on standard datasets and the results obtained show superiority of the proposed methods over single level dimensionality reduction techniques.

Evaluation of Three Feature Dimension Reduction Techniques for Machine Learning-Based Crop Yield Prediction Models

“…Filter methods have a low computational cost and are independent of the learning method [6,48,49]. However, they lack robustness against relationships among elements and element redundancy [48,49], and it is unclear how to choose the cut-off point for rankings to determine only dominant features [6].…”

“…Filter methods have a low computational cost and are independent of the learning method [6,48,49]. However, they lack robustness against relationships among elements and element redundancy [48,49], and it is unclear how to choose the cut-off point for rankings to determine only dominant features [6]. Generally, wrapper techniques outperform filter techniques [6,48,49] as they consider the feature dependencies and their collective contribution to model generation [6].…”

A Fuzzy Classifier with Feature Selection Based on the Gravitational Search Algorithm

“…The advantage of the wrapper methods is that they work together with the specific classification algorithm and account for the synergy of the joint usage of selected features. The disadvantages of the wrapper methods are the higher risk of overtraining and long time required to calculate classification accuracy [34].…”

An Efficient Feature Selection Model for IGBO Text

“…According to Pradnya and Manisha, 90% of the world's data is unstructured (textual data) and there is necessity for intelligent text analysis on the data [1]. The major challenge of intelligent text-based system is accuracy of the system and high dimensionality of feature space [2]. It is very important to use feature selection model to capture these challenges, by reducing high dimensionality of data for effective text-based system.…”