| Περίληψη: | The subject’s study of this specific thesis is the study, the analysis and the comparison of a multitude of variable selection techniques through an experimental evaluation on 10 conventional data sets.
In Chapter 2, we present the categories of variable selection methods as well as their mathematical background.
Then in Chapter 3, we refer to the Preprocessing stage of the datasets. By imply this process, we achieve the handling of missing values and possible asymmetry, the maintenance of uniformity in the types of variables, the delineation of a sufficient number of instances of the target variables and the complete removal of duplicates, constant and quasi variables.
In Chapter 4, the 1st Cycle of Experiments is carried out, which includes the analysis of all 10 pre-processed data sets, by using machine learning algorithms, without the assistance of variable selection methods. While in the 2nd Experiment Round, we implement and evaluate the 19 variable selection methods, which they have been chosen from the 2nd Chapter, by comparing the results.
Finally, in the last Chapter we compare and present the results of the two experimental cycles and propose solutions for further study and application.
|
|---|
