Analysis of Microarray Data using Machine Learning Techniques on Scalable Platforms

Abstract

Microarray-based gene expression profiling has been emerged as an efficient technique for classification, diagnosis, prognosis, and treatment of cancer disease. Frequent changes in the behavior of this disease, generate a huge volume of data. The data retrieved from microarray cover its veracities, and the changes observed as time changes (velocity). Although, it is a type of high-dimensional data which has very large number of features rather than number of samples. Therefore, the analysis of microarray high-dimensional dataset in a short period is very much essential. It often contains huge number of data, only a fraction of which comprises significantly expressed genes. The identification of the precise and interesting genes which are responsible for the cause of cancer is imperative in microarray data analysis. Most of the existing schemes employ a two phase process such as feature selection/extraction followed by classification. Our investigation starts with the analysis of microarray data using kernel based classifiers followed by feature selection using statistical t-test. In this work, various kernel based classifiers like Extreme learning machine (ELM), Relevance vector machine (RVM), and a new proposed method called kernel fuzzy inference system (KFIS) are implemented. The proposed models are investigated using three microarray datasets like Leukemia, Breast and Ovarian cancer. Finally, the performance of these classifiers are measured and compared with Support vector machine (SVM). From the results, it is revealed that the proposed models are able to classify the datasets efficiently and the performance is comparable to the existing kernel based classifiers. As the data size increases, to handle and process these datasets becomes very bottleneck. Hence, a distributed and a scalable cluster like Hadoop is needed for storing (HDFS) and processing (MapReduce as well as Spark) the datasets in an efficient way. The next contribution in this thesis deals with the implementation of feature selection methods, which are able to process the data in a distributed manner. Various statistical tests like ANOVA, Kruskal-Wallis, and Friedman tests are implemented using MapReduce and Spark frameworks which are executed on the top of Hadoop cluster. The performance of these scalable models are measured and compared with the conventional system. From the results, it is observed that the proposed scalable models are very efficient to process data of larger dimensions (GBs, TBs, etc.), as it is not possible to process with the traditional implementation of those algorithms. After selecting the relevant features, the next contribution of this thesis is the scalable viii implementation of the proximal support vector machine classifier, which is an efficient variant of SVM. The proposed classifier is implemented on the two scalable frameworks like MapReduce and Spark and executed on the Hadoop cluster. The obtained results are compared with the results obtained using conventional system. From the results, it is observed that the scalable cluster is well suited for the Big data. Furthermore, it is concluded that Spark is more efficient than MapReduce due to its an intelligent way of handling the datasets through Resilient distributed dataset (RDD) as well as in-memory processing and conventional system to analyze the Big datasets. Therefore, the next contribution of the thesis is the implementation of various scalable classifiers base on Spark. In this work various classifiers like, Logistic regression (LR), Support vector machine (SVM), Naive Bayes (NB), K-Nearest Neighbor (KNN), Artificial Neural Network (ANN), and Radial basis function network (RBFN) with two variants hybrid and gradient descent learning algorithms are proposed and implemented using Spark framework. The proposed scalable models are executed on Hadoop cluster as well as conventional system and the results are investigated. From the obtained results, it is observed that the execution of the scalable algorithms are very efficient than conventional system for processing the Big datasets. The efficacy of the proposed scalable algorithms to handle Big datasets are investigated and compared with the conventional system (where data are not distributed, kept on standalone machine and processed in a traditional manner). The comparative analysis shows that the scalable algorithms are very efficient to process Big datasets on Hadoop cluster rather than the conventional system.