S. Sasikala

A novel memetic algorithm for discovering knowledge in binary and multi class predictions based on support vector machine

Display Omitted Feature selection is important factor that hurtle classification accuracy. Also stands a back bone for the Dimensionality Reduction to boost the classification accuracy.To assure this requirement, a suitable feature selector is desired to be enhanced.This paper presents a novel memetic based feature selection model named Shapley Value Embedded Genetic Algorithm (SVEGA) Feature Selector to solve these multi objective feature selection responsibilities.The fitness value of an each feature subset is measured by combining the genetic algorithm with the shapely value measures to predict the prominent features and these measures are evaluated using Support Vector Machine (SVM) with the specific choice of kernel specification based on both binary and Multi class problems.The fitness function optimises the specificity and sensitivity of the model and achieves higher prediction accuracy with fewer number of features. In classification, every feature of the data set is an important contributor towards prediction accuracy and affects the model building cost. To extract the priority features for prediction, a suitable feature selector is schemed. This paper proposes a novel memetic based feature selection model named Shapely Value Embedded Genetic Algorithm (SVEGA). The relevance of each feature towards prediction is measured by assembling genetic algorithms with shapely value measures retrieved from SVEGA. The obtained results are then evaluated using Support Vector Machine (SVM) with different kernel configurations on 11+11 benchmark datasets (both binary class and multi class). Eventually, a contrasting analysis is done between SVEGA-SVM and other existing feature selection models. The experimental results with the proposed setup provides robust outcome; hence proving it to be an efficient approach for discovering knowledge via feature selection with improved classification accuracy compared to conventional methods.

A novel adaptive feature selector for supervised classification

Optimal feature Selection is an imperative area of research in medical data mining systems. Feature selection is an important factor that boosts-up the classification accuracy. In this paper we have proposed a adaptive feature selector based on game theory and optimization approach for an investigation on the improvement of the detection accuracy and optimal feature subset selection. Particularly, the embedded Shapely Value includes two memetic operators namely include and remove features (or genes) to realize the genetic algorithm (GA) solution. The use of GA for feature selection facilitates quick improvement in the solution through a fine tune search. An extensive experimental comparison on 22 benchmark datasets (both synthetic and microarray) from UCI Machine Learning repository and Kent ridge repository of proposed method and other conventional learning methods such as Support Vector Machine (SVM), Naive Bayes (NB), K-Nearest Neighbor (KNN), J48 (C4.5) and Artificial Neural Network (ANN) confirms that the proposed SVEGA strategy is effective and efficient in removing irrelevant and redundant features. We provide representative methods from each of wrapper, filter, conventional GA and show that this novel memetic algorithm - SVEGA yields overall promising results in terms of the evaluation criteria namely classification accuracy, number of selected genes, running time and other metrics over conventional feature selection methods.

Data classification using PCA based on Effective Variance Coverage (EVC)

Classification analysis of Medical diseases diagnosis has been performed extensively to find out the biological features and to differentiate intimately related diseases types that usually appear in the diagnosis of diseases. Many algorithms and techniques have been developed for the medical diseases classification process. These developed techniques accomplish feature based classification process with the aid of two basic phases namely dimensionality reduction through feature extraction, dimensionality reduction through feature selection. Among various dimensionality reduction techniques, this paper proposed prescribed statistical procedures to efficiently perform the classification process through feature extraction especially using PCA. To further substantiate and to analyze the performance, we conduct a deep study in Principle Component Analysis (PCA). The dimensionality reduction techniques perform the reduction in features through feature extraction and perform data classification with high accuracy. Based on the obtained results, we conclude that the performance study of PCA based on its Variance Coverage Range (VCR) over the several medical data sets work well. The study results that the statistical approach with PCA outperforms the classification performance.

RF-SEA-Based Feature Selection for Data Classification in Medical Domain

Dimensionality reduction is an essential problem in data analysis that has received a significant amount of attention from several disciplines. It includes two types of methods, i.e., feature extraction and feature selection. In this paper, we introduce a simple method for supervised feature selection for data classification tasks. The proposed hybrid feature selection mechanism (HFS), i.e., RF-SEA (ReliefF-Shapley ensemble analysis) which combines both filter and wrapper models for dimension reduction. In the first stage, we use the filter model to rank the features by the ReliefF(RF) between classes and then choose the highest relevant features to the classes with the help of the threshold. In the second stage, we use Shapley ensemble algorithm to evaluate the contribution of features to the classification task in the ranked feature subset and principal component analysis (PCA) is carried out as preprocessing step before both the steps. Experiments with several medical datasets proves that our proposed approach is capable of detecting completely irrelevant features and remove redundant features without significantly hurting the performance of the classification algorithm and also experimental results show obviously that the RF-SEA method can obtain better classification performance than singly Shapley-value-based or ReliefF (RF)-algorithm based method.

Generation of Cancelable Iris Template Using Bi-level Transformation

Cancelable biometric system is a transformation technique for securing biometric templates. This work proposes application of bi-level template securing technique at feature-level, and generates revocable templates. The bi-level transformation includes Discrete Fourier Transform and partial Hadamard based transformations on iris features, using user-specific key. The proposed bi-level transformation applied at feature-level provides better robustness and security against correlation attacks. A comprehensive analysis has been performed on the proposed approach to study the non-invertibility, diversity, revocability and matching performance on iris samples. The experimental results show that the proposed approach is promising, and deliver good performance.

Improving detection performance of artificial neural network by Shapley value embedded genetic feature selector

Abstract: This work is motivated by the interest in feature selection that greatly affects the detection accuracy of a classifier. The goals of this paper are (i) identifying optimal feature subset using a novel wrapper based feature selection algorithm called Shapley Value Embedded Genetic Algorithm (SVEGA), (ii) showing the improvement in the detection accuracy of the Artificial Neural Network (ANN) classifier with the optimal features selected, (iii) evaluating the performance of proposed SVEGA-ANN model on the medical datasets. The medical diagnosis system has been built using a wrapper based feature selection algorithm that attempts to maximize the specificity and sensitivity (in turn the accuracy) as well as by employing an ANN for classification. Two memetic operators namely “include” and “remove” features (or genes) are introduced to realize the genetic algorithm (GA) solution. The use of GA for feature selection facilitates quick improvement in the solution through a fine tune search. An extensive experimental evaluation of the proposed SVEGA-ANN method on 26 benchmark datasets from UCI Machine Learning repository and Kent ridge repository, with three conventional classifiers, outperforms state-of-the-art systems in terms of classification accuracy, number of selected features and running time.

Performance Tuning of PCA by CFS-Shapley Ensemble and Its Application to Medical Diagnosis

Selection of optimal features is an important area of research in medical data mining systems. Principal component analysis (PCA) is one among the most popular feature selection methods. Still PCA faces a drawback – i.e., the measurements from all of the original features are used in the projection to the lower dimensional space. Hence this work is aimed to tune the performance of PCA and classify the medical profiles. The proposed method is realized as an ensemble procedure with three steps – (i) feature selection using PCA, (ii) feature ranking with CFS and (iii) dimension reduction using Shapley Values Analysis. The variance coverage parameter of PCA is adjusted so as to yield maximum accuracy which are measured with specificity, sensitivity, precision and recall. This facilitates the selection of a compact set of superior features with uncompromised detection rates, remarkably at a low cost. To appraise the success of the proposed method, experiments were conducted across 6 different medical data sets using J48 decision tree classifier, which showed that the proposed procedure improves the classification efficiency and accuracy compared with individual usage.

A predictive model using improved Normalized Point Wise Mutual Information (INPMI)

In machine learning, selection of optimal features for the classifier is a critical problem. In order to address this problem a novel feature selection method named “Improved Normalized Point wise Mutual Information (INPMI)” is proposed. The proposed INPMI method coupled with Sequential forward search (SFS) finds the best feature subset to aid feature selection process. The key properties of evaluating feature subset i.e. relevancy and redundancy are analysed well. The classifiers like Naive Bayes, Support Vector Machine and J48 are used to determine the accuracy for the choice of features selected. Experimental results with benchmark medical datasets from UCI (University of California Irvine) machine learning database show that proposed INPMI-NB model with SFS, INPMI-SVM model with SFS, INPMI-J48model with SFS achieves 98.36 %, 98.90 %, 94.53 % classification accuracy and selects 22 features for erythemato-squamous diseases. Also the proposed work is evaluated on a World Aircraft dataset to prove its generalization ability. Experimental results prove that the proposed INPMI method outperforms the existing systems.

An efficient feature selection paradigm using PCA-CFS-Shapley values ensemble applied to small medical data sets

The precise diagnosis of patient profiles into categories, such as presence or absence of a particular disease along with its level of severity, remains to be a crucial challenge in biomedical field. This process is realized by the performance of the classifier by using a supervised training set with labeled samples. Then based on the result obtained, the classifier is allowed to predict the labels of new samples. Due to presence of irrelevant features it is difficult for standard classifiers from obtaining good detection rates. Hence it is important to select the features which are more relevant and by with good classifiers could be constructed to obtain a good accuracy and efficiency. This study is aimed to classify the medical profiles, and is realized by feature extraction (FE), feature ranking (FR) and dimension reduction methods (Shapley Values Analysis) as a hybrid procedure to improve the classification efficiency and accuracy. To appraise the success of the proposed method, experiments were conducted across 6 different medical data sets using J48 decision tree classifier. The experimental results showed that using the PCA-CFS-Shapley Values analysis procedure improves the classification efficiency and accuracy compared with individual usage.