Feature selection in pathology detection using hybrid multidimensional analysis

Area of Science:

• Computational linguistics
• Biomedical engineering
• Statistical analysis

Background:

• Heuristic algorithms reduce computational complexity but require effective cost functions.
• Existing methods like Principal Component Analysis (PCA) focus on data variance, not class separability.
• Multivariate statistical methods offer potential for improved feature selection.

Purpose of the Study:

• To propose a novel heuristic feature selection algorithm utilizing Multivariate Analysis of Variance (MANOVA) as a cost function.
• To evaluate the algorithm's effectiveness in classifying hypernasal versus normal voices in patients with Cleft Lip and/or Palate (CLP).
• To compare the proposed method against an alternative feature selection approach based on univariate and bivariate analysis.

Main Methods:

• Development of a heuristic search algorithm incorporating MANOVA as the cost function for feature selection.
• Application of the algorithm to a dataset of voice recordings from CLP patients.
• Comparative analysis of classification performance, computational time, and feature reduction ratio against a baseline method.

Main Results:

• The MANOVA-based feature selection algorithm demonstrated superior classification performance.
• The proposed method achieved a significant reduction in feature dimensionality.
• The heuristic approach using MANOVA proved computationally efficient compared to the alternative.

Conclusions:

• MANOVA is an effective cost function for heuristic feature selection in voice analysis for CLP patients.
• The proposed algorithm offers a promising approach for improving diagnostic accuracy and efficiency in speech pathology.
• This method provides a more class-discriminatory feature subspace than variance-based techniques like PCA.