الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 117 |  |  |
| | | from | 117 | | |
Abstract
Microarray technology has become one of the most relevant experimental developments in molecular biology in recent years. Microarrays have allowed scientists to study the expression of thousands of genes simultaneously. However, the large number of genes, the small number of samples, the high level of noise, the diversity of data analysis methods available and the complexity of biological systems has significantly complicated the analysis of such datasets. Due to the challenging nature of this topic, it has drawn much attention from the data mining community. Most of common attention perspectives are feature subset selection and classification. In the last decade, feature selection technique has become an important tool for lots of bioinformatics applications by microarray data, such as cancer classification, biologic network inference, and expression correlation analysis and disease biomarker identification. With the development of genome research, especially with the wide use of the high-throughput microarray chip technology, researchers have obtained a large number of microarray related data and information at the gene level. Most of the gene expression datasets have small number of samples and tens of thousands of genes. Besides, classification techniques which represent another attention perspective including J48, naïve Bayes, support vector machine, deep learning …etc. In this thesis, we designed two embedded feature subsets selection using naïve Bayes evaluation where everyone is considered as a model. First is genetic algorithm hybrid naïve Bayes while second is binary gray wolf optimization guided by naïve Bayes. On the top of each one, a deep learning model exist which is based on Boltzmann neural network. The proposed DBN-DNN is able to adapt for small number of samples per gene expression datasets.