Document Details

Document Type : Thesis 
Document Title :
OPTIMIZED FEATURE SELECTION TECHNIQUES BASED ON HIGH PERFORMANCE COMPUTING
تقنية اختيار الخصائص المثالية المبنية على الحوسبة عالية الأداء
 
Subject : Faculty of Computing and Information Technology 
Document Language : Arabic 
Abstract : The Branch and bound (BB) algorithm undergoes an exponential growth in feature selection as the number of features increases, which may require in the worst cases, exploring the whole tree looking for an optimal solution. This research presents an enhancement in the BB algorithm for feature selection using an approximate monotonic criteria function and High Performance Computing (HPC). The enhanced parallel sub-optimal proposed version of the BB algorithm seeking for the solution by cutting off unpromising paths and deleting multiple features at each internal tree node using tau variable. The number of deleted features in each node is determined by the tau variable value, which best specified according to the correlation between the features. KNN (K-Nearest Neighbors) classifier is used as the J function that computes the highest accuracy of the subset of features. The experiment was applied to different datasets and compared to the original BB algorithm and numerous selection methods. Although the proposed algorithm does not guarantee the selection of the optimal features subset, it is able to reach a high accuracy comparable to the optimal version of the original BB algorithm, and perhaps better than it. The results show promising results in terms of accuracy, elapsed time, and tree size. 
Supervisor : Dr. Mohamed Dahab 
Thesis Type : Master Thesis 
Publishing Year : 1442 AH
2020 AD 
Added Date : Saturday, September 12, 2020 

Researchers

Researcher Name (Arabic)Researcher Name (English)Researcher TypeDr GradeEmail
عهود نايف الحربيAlharbi, Ahood NaifResearcherMaster 

Files

File NameTypeDescription
 46737.pdf pdf 

Back To Researches Page