P. K. Nizar Banu

Rough Set Based Feature Selection for Egyptian Neonatal Jaundice

This paper analyses rough set based feature selection methods for early intervention and prevention of neurological dysfunction and kernicterus that are the major causes of neonatal jaundice. Newborn babies develop some degree of jaundice which requires high medical attention. Improper prediction of diseases may lead to choose unsuitable type of treatment. Traditional rough set based feature selection methods and tolerance rough set based feature selection methods for supervised and unsupervised approach is applied for Egyptian neonatal jaundice dataset. Features responsible for prediction of Egyptian neonatal jaundice is analyzed using supervised quick reduct, supervised entropy based reduct and Unsupervised Tolerance Rough Set based Quick Reduct (U-TRS-QR). Results obtained demonstrate features selected by U-TRS-QR are highly accurate and will be helpful for physicians for early diagnosis.

Gene Clustering Using Metaheuristic Optimization Algorithms

Gene clustering is a familiar step in the exploratory analysis of high dimensional biological data. It is the process of grouping genes of similar patterns in the same cluster and aims at analyzing the functions of gene that leads to the development of drugs and early diagnosis of diseases. In the recent years, much research has been proposed using nature inspired meta-heuristic algorithms. Cuckoo Search is one such optimization algorithm inspired from nature by breeding strategy of parasitic bird, the cuckoo. This paper proposes cuckoo search clustering and clustering using levy flight cuckoo search for grouping brain tumor gene expression dataset. A comparative study is made with genetic algorithm, PSO clustering, cuckoo search clustering and clustering using levy flight cuckoo search. Levy flight is an important property of levy distribution which covers the entire search space. Breeding pattern of cuckoo is associated with the genes that cause tumor to grow and affect other organs gradually. Clusters generated by these algorithms are validated to find the closeness among the genes in a cluster and separation of genes between clusters. Experimental results carried out in this paper show that cuckoo search clustering outperforms other clustering methods used for experimentation.

Unsupervised hybrid PSO - Quick reduct approach for feature reduction

Feature reduction reduces the dimensionality of a database and selects more informative features by removing the irrelevant features. Selecting features in unsupervised learning scenarios is a harder problem than supervised feature selection due to the absence of class labels that would guide the search for relevant features. PSO is an evolutionary computation technique which finds global optimum solution in many applications. Rough set is a powerful tool for data reduction based on dependency between attributes. This work combines the benefits of both PSO and rough sets. This paper describes a novel Unsupervised PSO based Quick Reduct (US-PSO-QR) for feature selection which employs a population of particles existing within a multi-dimensional space. The performance of the proposed algorithm is compared with the existing unsupervised feature selection methods and the efficiency is measured by using K-Means Clustering and Rough K-Means Clustering.

Performance Analysis of Hard and Soft Clustering Approaches For Gene Expression Data

Mining gene expression data is growing rapidly to predict gene expression patterns and assist clinicians in early diagnosis of tumor formation. Clustering gene expression data is the most important phase, helps in finding group of genes that are highly expressed and suppressed. This paper analyses the performance of most representative hard and soft off-line clustering algorithms: K-Means, Fuzzy C-Means, Self Organizing Maps SOM based clustering and Genetic Algorithm GA based clustering for brain tumor gene expression dataset. Clusters produced by the clustering algorithms are the indications of the cellular processes. Clustering results are evaluated using clustering indices such as Xie-Beni index XB, Davies-Bouldin index DB, Mean Absolute Error MAE, Root Mean Squared Error RMSE and Dunns Index DI along with the time taken to find the compactness and separation of clusters. Experimental results prove soft clustering approaches works well to predict clusters of highly expressed and suppressed genes.

Unsupervised hybrid PSO — Relative reduct approach for feature reduction

Feature reduction selects more informative features and reduces the dimensionality of a database by removing the irrelevant features. Selecting features in unsupervised learning scenarios is a harder problem than supervised feature selection due to the absence of class labels that would guide the search for relevant features. Rough set is proved to be efficient tool for feature reduction and needs no additional information. PSO (Particle Swarm Optimization) is an evolutionary computation technique which finds global optimum solution in many applications. This work combines the benefits of both PSO and rough sets for better data reduction. This paper describes a novel Unsupervised PSO based Relative Reduct (US-PSO-RR) for feature selection which employs a population of particles existing within a multi-dimensional space and dependency measure. The performance of the proposed algorithm is compared with the existing unsupervised feature selection methods USQR (UnSupervised Quick Reduct) and USSR (UnSupervised Relative Reduct) and the effectiveness of the proposed approach is measured by using Clustering evaluation indices.

Informative Gene Selection - An evolutionary approach

Feature selection is one of the major challenges in the analysis of gene expression data, as the number of genes significantly exceeds the number of samples. Principal Component Analysis (PCA), one of the most popular dimensionality reduction techniques, reveals the underlying factors or combinations of original variables without any information loss. This paper studies the application of PCA based on Eigen vectors of covariance and Singular Value Decomposition (SVD) for gene expression dataset as well and explores the problem of feature subset selection, by selecting highly dominating genes to predict cancer at an early stage. The proposed Informative Gene Selection method aims to identify a subset of genes with higher accuracy to represent original genes. Computational time and clustering accuracy is also recorded separately. The proposed method results with more interpretable features that help to identify the target disease quickly. The prominent results show the effectiveness of the proposed algorithm.

Fuzzy firefly clustering for tumour and cancer analysis

Swarm intelligence represents a meta-heuristic approach to solve a wide variety of problems. Searching for similar patterns of genes is becoming very essential to predict the expression of genes under various conditions. Firefly clustering inspired by the behaviour of fireflies helps in grouping genes that behave alike. Contrasting hard clustering methodology, fuzzy clustering assigns membership values for every gene and predicts the possibility of belonging to every cluster. To distinguish highly expressed and suppressed genes, the research in this paper proposes an efficient fuzzy-firefly clustering by integrating the merits of firefly and fuzzy clustering. The proposed method is compared with other swarm optimisation based clustering algorithms. It is applied on five gene expression datasets. The clusters resulting from the proposed algorithm provide interpretations of different gene expression patterns present in the cancer datasets. Experimental results show the excellent performance of fuzzy-firefly clustering to separate co-expressed and co-regulated genes.

Evaluation of Fitness Functions for Swarm Clustering Applied to Gene Expression Data

Clustering problem is being studied by many of the researchers using swarm intelligence. However, the search space is not carried out entirely randomly; a proper fitness function is required to determine the next step in the search space. This paper studies Particle Swarm Optimization (PSO) based clustering with two different fitness functions namely Xie-Beni and Davies-Bouldin indices for brain tumor gene expression dataset. Clustering results are validated using Mean Absolute Error (MAE) and Dunn Index (DI). To analyze function of genes, genes that have similar expression patterns should be grouped and the datasets should be presented to the physicians in a meaningful way. High usability of algorithm and the encouraging results suggests that swarm clustering (PSO based clustering) with Davies-Bouldin index as fitness functions with respect to Dunn index can be a practical tool for analyzing gene expression patterns.

A Comparative Analysis of Rough Set Based Intelligent Techniques for Unsupervised Gene Selection

As the micro array databases increases in dimension and results in complexity, identifying the most informative genes is a challenging task. Such difficulty is often related to the huge number of genes with very few samples. Research in medical data mining addresses this problem by applying techniques from data mining and machine learning to the micro array datasets. In this paper Unsupervised Tolerance Rough Set based Quick Reduct U-TRS-QR, a diverse feature selection algorithm, which extends the existing equivalent rough sets for unsupervised learning, is proposed. Genes selected by the proposed method leads to a considerably improved class predictions in wide experiments on two gene expression datasets: Brain Tumor and Colon Cancer. The results indicate consistent improvement among 12 classifiers.

Harmony Search PSO Clustering for Tumor and Cancer Gene Expression Dataset

Enormous quantity of gene expression data from diverse data sources are accumulated due to the modern advancement in microarray technology that leads to major computational challenges. The foremost step towards addressing this challenge is to cluster genes which reveal hidden gene expression patterns and natural structures to find the interesting patterns from the underlying data that in turn helps in disease diagnosis and drug development. Particle Swarm Optimization (PSO) technique is extensively used for many practical applications but fails in finding the initial seeds to generate clusters and thus reduces the clustering accuracy. One of the meta-heuristic optimization algorithms called Harmony Search is free from divergence and helps to find out the near-global optimal solutions by searching the entire solution space. This paper proposes a novel Harmony Search Particle Swarm Optimization (HSPSO) clustering algorithm and is applied for Brain Tumor, Colon Cancer, Leukemia Cancer and Lung Cancer gene expression datasets for clustering. Experimental results show that the proposed algorithm produces clusters with better compactness and accuracy, in comparison with K-means clustering, PSO clustering (swarm clustering) and Fuzzy PSO clustering.