Research Area:  Machine Learning
This research focuses mainly on the binary class imbalance problem in data mining. It investigates the use of combined approaches of data and algorithmic level solutions. Moreover, it examines the use of swarm intelligence and population-based techniques to combat the class imbalance problem at all levels, including at the data, algorithmic, and feature level. It also introduces various solutions to the class imbalance problem, in which swarm intelligence techniques like Stochastic Diffusion Search (SDS) and Dispersive Flies Optimisation (DFO) are used. The algorithms were evaluated using experiments on imbalanced datasets, in which the Support Vector Machine (SVM) was used as a classifier. SDS was used to perform informed undersampling of the majority class to balance the dataset. The results indicate that this algorithm improves the classifier performance and can be used on imbalanced datasets. Moreover, SDS was extended further to perform feature selection on high dimensional datasets. Experimental results show that SDS can be used to perform feature selection and improve the classifier performance on imbalanced datasets. Further experiments evaluated DFO as an algorithmic level solution to optimise the SVM kernel parameters when learning from imbalanced datasets. Based on the promising results of DFO in these experiments, the novel approach was extended further to provide a hybrid algorithm that simultaneously optimises the kernel parameters and performs feature selection.
Name of the Researcher:  Alhakbani, Haya
Name of the Supervisor(s):  Dr. Mohammad AL-RIFAIE
Year of Completion:  2019
University:  University of London
Thesis Link:   Home Page Url