[Paper Review] A Harmony Search Based Wrapper Feature Selection Method for Holistic Bangla word Recognition

Supratim Das, Pawan Kumar Singh+3 more|arXiv (Cornell University)|Jul 26, 2017

Text and Document Classification Technologies13 references3 citations

TL;DR

This paper proposes a Harmony Search (HS) based wrapper feature selection method to reduce dimensionality in holistic Bangla handwritten word recognition. By selecting 48 optimal features from an initial set of 65 elliptical features, the method achieves 90.29% recognition accuracy using a Multi Layer Perceptron classifier—surpassing previous results and outperforming GA, PSO, and PCA in both accuracy and feature efficiency.

ABSTRACT

A lot of search approaches have been explored for the selection of features in pattern classification domain in order to discover significant subset of the features which produces better accuracy. In this paper, we introduced a Harmony Search (HS) algorithm based feature selection method for feature dimensionality reduction in handwritten Bangla word recognition problem. This algorithm has been implemented to reduce the feature dimensionality of a technique described in one of our previous papers by S. Bhowmik et al.[1]. In the said paper, a set of 65 elliptical features were computed for handwritten Bangla word recognition purpose and a recognition accuracy of 81.37% was achieved using Multi Layer Perceptron (MLP) classifier. In the present work, a subset containing 48 features (approximately 75% of said feature vector) has been selected by HS based wrapper feature selection method which produces an accuracy rate of 90.29%. Reasonable outcomes also validates that the introduced algorithm utilizes optimal number of features while showing higher classification accuracies when compared to two standard evolutionary algorithms like Genetic Algorithm (GA), Particle Swarm Optimization (PSO) and statistical feature dimensionality reduction technique like Principal Component Analysis (PCA). This confirms the suitability of HS algorithm to the holistic handwritten word recognition problem.

Motivation & Objective

• To address high-dimensional feature space in holistic Bangla word recognition that hinders classification efficiency.
• To reduce feature dimensionality while preserving or improving recognition accuracy.
• To evaluate the effectiveness of Harmony Search (HS) as a wrapper feature selection method in comparison to established evolutionary algorithms and statistical techniques.
• To identify a minimal yet optimal feature subset that maximizes classification performance for handwritten Bangla words.

Proposed method

• The Harmony Search (HS) algorithm is employed as a wrapper method to search for the optimal subset of features from an initial set of 65 elliptical features.
• The HS algorithm iteratively generates new feature combinations (harmony vectors), evaluates them using a Multi Layer Perceptron (MLP) classifier, and updates the harmony memory based on classification accuracy.
• Fitness evaluation is performed using the recognition accuracy of the MLP classifier on a validation set, guiding the search toward higher-performing feature subsets.
• The algorithm maintains a harmony memory of candidate solutions and uses pitch adjustment and bandwidth parameters to explore and exploit the search space effectively.
• The process continues until a predefined number of iterations or convergence criteria are met, yielding the best-performing feature subset.
• The selected feature subset is then validated on a test set to measure final recognition accuracy.

Experimental results

Research questions

• RQ1 Can the Harmony Search algorithm effectively reduce feature dimensionality in holistic Bangla word recognition while maintaining or improving classification accuracy?
• RQ2 How does the HS-based wrapper feature selection method compare to Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Principal Component Analysis (PCA) in terms of recognition accuracy and feature reduction?
• RQ3 What is the optimal number of features required to achieve high recognition accuracy in the holistic Bangla word recognition task?
• RQ4 Does the HS-based method select a more compact and informative feature subset than standard dimensionality reduction techniques?

Key findings

• The proposed HS-based wrapper feature selection method achieved a recognition accuracy of 90.29% using only 48 features, representing a 75% reduction from the original 65-feature set.
• The method significantly outperformed the baseline approach (81.37% accuracy with 65 features), demonstrating the benefit of intelligent feature subset selection.
• The HS-based method surpassed both Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) in classification accuracy, confirming its superiority in this context.
• The method also outperformed Principal Component Analysis (PCA), a statistical dimensionality reduction technique, in terms of recognition accuracy.
• The results validate that the HS algorithm is well-suited for feature selection in holistic handwritten Bangla word recognition due to its ability to balance exploration and exploitation.
• The study confirms that optimal feature selection leads to higher accuracy with fewer features, enhancing computational efficiency without sacrificing performance.