Enhanced Ant Colony Algorithm for Best Features Selection for a Decision Tree Classification of Medical Data

Enhanced Ant Colony Algorithm for Best Features Selection for a Decision Tree Classification of Medical Data

Abdiya Alaoui (Djillali Liabes University Sidi Belabbes, Algeria) and Zakaria Elberrichi (Djillali Liabes University Sidi Belabbes, Algeria)
Copyright: © 2020 |Pages: 16
DOI: 10.4018/978-1-7998-1021-6.ch015

Abstract

Classification algorithms are widely applied in medical domain to classify the data for diagnosis. The datasets have considerable irrelevant attributes. Diagnosis of the diseases is costly because many tests are required to predict a disease. Feature selection is one of the significant tasks of the preprocessing phase for the data. It can extract a subset of attributes from a large set and exclude redundant, irrelevant, or noisy attributes. The authors can decrease the cost of diagnosis by avoiding numerous tests by selection of features, which are important for prediction of disease. Applied to the task of supervised classification, the authors construct a robust learning model for disease prediction. The search for a subset of features is an NP-hard problem, which can be solved by the metaheuristics. In this chapter, a wrapper approach by hybridization between ant colony algorithm and adaboost with decision trees to ameliorate the classification is proposed. The authors use an enhanced global pheromone updating rule. With the experimental results, this approach gives good results.
Chapter Preview
Top

Introduction

One important task of data mining is the supervised classification, which allows predicting if an instance will be assigned to a predefined class. It has many human applications: it is especially appropriate for automated decision making problems such as medical diagnosis. For some application areas, it is essential to produce the explicit classification procedures for the user. For example a decision trees for a medical diagnosis. The medical data is with many Features, diagnosis of the diseases is expensive as many tests are required to predict the disease.

The computational complexity of a classification algorithm may suffer from the curse of dimensionality caused by numerous features. Often a dataset has too many irrelevant features. Selecting only the most important ones can only be beneficial. Dimensionality reduction plays an important role in the domain of medicine as it contains many attributes.

Feature selection is often one of the imperative tasks in data preprocessing for data mining that has been used in different domains, it used to solve many problems in Bioinformatics (Dif & Elberrichi, 2019; Fan, Tang, Tian, & Wu, 2019), opinion classification (Sangam & Shinde, 2019), text categorization (Tang, Dai, & Xiang, 2019), cancer Classification (Lindqvist & Price, 2018) and sentiment analysis (Zheng, Wang, & Gao, 2018).

It consists in removing irrelevant, redundant, and noisy data. It reduces automatically the number of features, and has an immediate effect for the applications: speeding up the data mining algorithm, improving the model performance such as predictive accuracy.

Many problems connected to feature selection have been shown to be NP-Hard (Cotta & Moscato, 2003) and can be solved by using metaheuristics. Many optimization techniques such as genetic algorithms (Feng, Xuezheng, Yanqing, & Bourgeois, 2007), Tabu Search (Hongbin & Guangyu, 2002), Simulated Annealing (Ronen & Jacob, 2004) and Particle swarm optimization (Bing, Mengjie, & Will, 2014; Liam, Bing, Mengjie& Lin, 2012) have been used for solving the feature selection problem. Among these approaches, the Ant Colony Optimization (ACO) has been explored significantly to solving these problems. It is thought to be a promising method for a good selection of features (Al-Ani, 2005; Kumar, 2011; Shahzad, 2010).

Many approaches have been proposed in the literature for feature selection problem in medical domain, some approaches are shown in Table 1. Where the metaheuristics are used to determine the optimal feature subset with improved classification accuracy in disease diagnosis.

Feature selection methods are Filter and wrapper based on evolutionary and swarm algorithms.

Table 1.
Filter and Wrapper Feature selection methods.
          References          Algorithms          Type          Datasets
          (Subanya & Rajalaxmi, 2014)          Binary Artificial Bee Colony/ K–Nearest Neighbor (BABC k–NN)          Wrapper          Heart disease
          (Harb & Desuky, 2014)          Particle Swarm Optimization (PSO)          Filter and Wrapper          Breast Cancer, Heart Statlog, Dermatology
          (Ravi et al., 2014)          Genetic algorithm /SVM          Wrapper          Wisconsin Breast, cancer Pima Diabetes, Dermatology, Thoracic Surgery, Mammographic,
          (Brezocnik et al., 2018)          Swarm Intelligence Algorithms, Review          Filter, Wrapper and embedded          Medical Data and Other
          (Darwish et al., 2018)          bio-inspired algorithms          Wrapper          breast cancer (WBCD and WPBC)
          (Nematzadeh et al., 2019)          whale algorithm with Mutual Congestion          Filter          Colon

Key Terms in this Chapter

Supervised Classifier: Allows predicting if an instance will be assigned to a predefined class to classify the data for diagnosis and prognosis.

Feature Selection: By this task removing irrelevant, redundant, and noisy data.

Medical Datasets: Set of instances for diagnosis and prognosis.

Improved Ant Colony Optimization: Ant colony optimization with improved global pheromone updating rule.

Best Fitness: Best accuracy of best subset found.

Global Best Subset: Selecting only the most important ones can only be beneficial.

Relevant Features: Helpful attributes for diagnosis and thus can’t be removed before learning.

Complete Chapter List

Search this Book:
Reset