Thesis (Ph.D.) - Cairo University - Faculty of Computers and Artificial Intelligence - Department of Computer Science

Cancer is one of the most common life-threatening diseases. There is remarkably rapid development in producing cancer genomic datasets, especially the microarray datasets. Despite the abundance of data in the microarray datasets, they have a drawback, which is dimensionality. In addition to the multi-omic datasets, which combine more than one biological dataset for the same patients set, each is related to a specific type of biological features. The insights we gain for a particular biological problem or disease are double using multi-omic rather than using one dataset. It is like investigating a problem from many dimensions rather than using one dimension. On the other hand, the difficulty of analysis is increased. Therefore, new special mathematical models are needed to deal with these new cancer datasets. In this thesis, we propose two methods to deal with these new cancer datasets. At first, we propose a new filter-based gene selection method that merges the Dragonfly algorithm and the correlation-based feature selection to reduce the redundancy between the genes selected and increase the relevance between the selected genes and the decision. The proposed method is compared with nine famous feature selection methods.The experiments are applied to five widely used public microarray datasets. The used evaluation criterion of the selected features is the average accuracy of classification using three different classifiers: support vector machine, naïve Bayes, and decision tree. Experimental results demonstrate that the proposed method is efficient and performs better than the other nine methods used in the experiment. It also shows that the proposed method can be used with any one of the three classifiers included in our study to obtain an efficient automatic cancer diagnostic system

Issued also as CD