الفهرس 
AbstractOnly 14 pages are availabe for public view
 |  | from | 66 |  |  |
| | | from | 66 | | |
Abstract
Feature selection is a very critical step in machine learning. It can affect the whole performance of a system from one case to the opposite. While the need for larger datasets increase, feature selection becomes a main process in building a machine learning model cycle. It starts with big datasets and by a feature selection algorithm only the most significant features remain in action and used to make a better model. Embedded methods as it is named embed both the feature selection step and the learning process by using the machine learning algorithm to select features during training. The common use of this method is neural networks and support vector machine. Although it seems time-consuming and tedious, but it gives a deep connection between the data and the problem that we use these data with, and accordingly in most cases a good accuracy of the learning model is gained. The last class is wrapper methods which use two different algorithms for both feature selection process and machine learning process like filter methods someway, but it uses the machine learning algorithm to evaluate the selected features with some iterations to come up with the best features that can be used to make a learning model for a specific problem. Wrapper methods use heuristics and biological inspired techniques to make the feature selection operation. In this work we focus on bio-inspired algorithms especially Grey Wolf Optimizer (GWO) which is considered a powerful optimization technique inspired from the way used by grey wolves to hunt. Such an algorithm that mimic the nature gives good results in solving many optimization problems including feature selection. The binary version of GWO (BGWO) uses binary values for wolves’ positions rather than probabilistic values in the original GWO. Combining a bio-computing algorithm with quantum operations is the main work of this thesis. The great effects of quantum computing in most computer science fields pushed us to discover how it will affect an operation like feature selection with a method like GWO. In this thesis we introduce a new perspective of GWO using a Quantum operation producing an enhanced quantum inspired BGWO for feature selection. Our method was evaluated against BGWO method by comparing the error rate, number of eliminated features and global optima iteration number. it showed a better accuracy and eliminated larger number of features with good performance.