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

In the process of selecting the most appropriate classification algorithm, there are two main tasks. The determination of the factors that will be used in the selection process and the methodology that will be used to make use of these factors and decide upon the most appropriate algorithm to solve the problem. The first task is to characterize datasets; by extracting its characteristics/meta-data or by adopting a reasonable characterization approach, whilst the second task is the learning and deciding task based on the characterization. Choosing the most appropriate classification algorithm for classification problem is becoming a strategically important task in the data-mining process. Due to the availability of numerous classification algorithms in the area of data mining for solving the same kind of problem, with little guidance available for recommending the most appropriate algorithm to use which gives best results for the dataset at hand, this task becomes more and more complicated. As a way of optimizing the chances of recommending the most appropriate classification algorithm for a dataset, a two-step study was conducted: (1) Survey study focusing on the different factors considered by data miners and researchers in different studies when manually selecting the classification algorithms that will yield desired knowledge for the dataset at hand, a categorization tree was created for the measurable factors.The categorization tree, groups and categorizes these factors so that they can be exploited by recommendation software tools. (2) Experimental study of an automated tool based on a pure Collaborative Filtering Recommender System, User-Item approach, to recommend the most appropriate classification algorithm for a classification problem, relying on historical datasets meta-data

Issued also as CD