Date of Submission
Master of Science in Computer Science (MSCS)
Machine Learning and Data Science
Genetic programming (GP), a capable machine learning and search method, motivated by Darwinian-evolution, is an evolutionary learning algorithm which automatically evolves computer programs in the form of trees to solve problems. This thesis studies the application of GP for data mining and image processing. Knowledge discovery and data mining have been widely used in business, healthcare, and scientific fields. In data mining, classification is supervised learning that identifies new patterns and maps the data to predefined targets. A GP based classifier is developed in order to perform these mappings. GP has been investigated in a series of studies to classify data; however, there are certain aspects which have not formerly been studied.
We propose an optimized GP classifier based on a combination of pruning subtrees and a new fitness function. An orthogonal least squares algorithm is also applied in the training phase to create a robust GP classifier. The proposed GP classifier is validated by 10-fold cross validation. Three areas were studied in this thesis. The first investigation resulted in an optimized genetic-programming-based classifier that directly solves multi-class classification problems. Instead of defining static thresholds as boundaries to differentiate between multiple labels, our work presents a method of classification where a GP system learns the relationships among experiential data and models them mathematically during the evolutionary process. Our approach has been assessed on six multiclass datasets. The second investigation was to develop a GP classifier to segment and detect brain tumors on magnetic resonance imaging (MRI) images. The findings indicated the high accuracy of brain tumor classification provided by our GP classifier. The results confirm the strong ability of the developed technique for complicated image classification problems. The third was to develop a hybrid system for multiclass imbalanced data classification using GP and SMOTE which was tested on satellite images. The finding showed that the proposed approach improves both training and test results when the SMOTE technique is incorporated. We compared our approach in terms of speed with previous GP algorithms as well. The analyzed results illustrate that the developed classifier produces a productive and rapid method for classification tasks that outperforms the previous methods for more challenging multiclass classification problems. We tested the approaches presented in this thesis on publicly available datasets, and images. The findings were statistically tested to conclude the robustness of the developed approaches.