الفهرس | Only 14 pages are availabe for public view |
Abstract The thesis tries to emphasize the power of genetic algorithm (GA) in dealing with symbolic data and supervised learning. Genetic algorithm tries to find all perfect solutions hidden in a hyper-space of a number of attributes. The thesis tries to find out an algorithm based on the genetic algorithm that solves for finding out hidden different concepts inside a number of given examples in a noisy environment. A set of derived Boolean and fuzzy attributes is defined to a number of examples belonging to a single table in a database file. The attributes then get their values from the database table using SQL SELECT statements. Then, a genetic algorithm and a classifier system try to extract hidden common concepts between these attributes. The concept structure is represented using a prefix Boolean expression grammar. This representation method can discover new hidden unexpected knowledge inside the examples. Both of the genetic algorithm and classifier system needs to be varied to cope with the new representation structure.In order to find a number of different solutions, the genetic algorithm has been modified to punish trivial, alike concepts. The punishment occurs in such a manner that enables only a single concept to maintain its power while the other concepts lose some of their fitness to encourage searching in different directions. The superset concepts also are punished when there are perfect subset concepts. That enables getting more specific concepts rather than general ones.The two proposed methods try to solve for two real examples (4-bit shift register and student-marks). Then, a source of noise is applied on both examples to simulate the effect of odd cases and noisy data on the input data and the methods run again to extract concepts under noise.Keywords Artificial Intelligence, Machine Learning, Concept Learning, Data Mining, Genetic Algorithm, Classifier Learning System, Supervised Learning, Fuzzy Logic. |