Article Preview
Top1. Introduction
The field of bioinformatics has for many years already attracted an excellent interest within the scientific community as a result of it disclose terribly made views for the understanding of biological phenomena. Problems addressed include genome sequencing, protein structure modeling, or the reconstruction of phylogenetic trees (phylogeny), attribute selection. These problems require collaboration between biologists and computer scientists because the problems to be dealt with often pose great algorithmic difficulties.
Nowadays, folks will get the expression datasets of thousands of genes at the same time victimization microarray technology. one among the vital fields in victimization these organic phenomenon datasets is to classify and predict the diagnostic class of a sample. Actually, precise diagnosing and classification is crucial for in the treatment of health problem (Mei, Shen, & Ye, 2009).
Microarray knowledge analysis are often administrated in keeping with a minimum of 2 totally different and complementary views. In one hand, knowledge cluster (non-supervised classification) aims to spot teams of genes or teams of experimental conditions that exhibit similar expression patterns. In such a context bi-clustering is especially fascinating since it permits the synchronous identification of teams of genes that show similar expression patterns across specific teams of experimental conditions (samples) (Duvaland & Hao, 1999).
The processing of these data, therefore, requires reducing the number of genes to provide a subset of relevant genes and construct a classifier predicting the type of tumor that characterizes a cell sample. The selection of attributes is a complex problem that has already been widely studied, but the dimensions of the biochip data require specific approaches (several thousand genes).
Several machine learning methods have been applied to gene expression data classification. Ben-Dor has proved that Nearest Neighbor Algorithm (NN Algorithm), Boosting Algorithm and clustering-based classification algorithm can all be applied to gene expression data classification (Ben-Dor et al., 2000). Support Vector Machines (SVMs), a classical machine learning algorithm was firstly applied to gene expression data classification by (Grundy et al., 2000). A principal component and discriminant analysis method of tumor classification was proposed by (Xiong, Jin & Li, 2000). Data mining is one of the important phases of knowledge data discovery, there is a technique which is used to find new, hidden and useful patterns of knowledge from large databases. There are several data mining methods such as Prediction, Clustering, and Classification (Lavanya and Usha Rani, 2011).
Indeed, the extremely rapid growth of the data collected in the databases and the need for effective responsiveness on the part of decision-makers to this new information have stimulated the rapid development of Knowledge Extraction in the last decade. From the Data (ECD). The ECD, or Knowledge Discovery in Databases, is a “non-trivial process of identifying unknown, valid and potentially exploitable structures in databases” [FPSS96]. The ECD consists of four phases (see Figure 1): data acquisition and storage (Data Warehousing), data pre-processing, data mining and post-processing, our work is mainly at the level of stage three: data mining. This step integrates both the choice of the appropriate modeling and the method to be used as well as its application to the search for underlying data structures and to the creation of explanatory and/or predictive models.
This step can be broken down into three major tasks: discrimination (supervised classification), categorization or clustering, and the search for association rules. These tasks can, by their formalization, be modeled in combinatorial optimization problems.
From the formalization of these various tasks of extraction of knowledge to problems of combinatorial optimization, it is necessary to implement methods of resolution. Usually, the methods used in this framework are of three types: exact methods, metaheuristics, and task-specific heuristic methods.
The selection of attributes is a complex problem that has already been widely studied, but the dimensions of the biochip data require specific approaches (several thousand genes).