Knowledge Discovery in Databases Processing Using Improved Data Mining Techniques

Abstract

This thesis focuses on problems related to supervised learning for data mining and system development for knowledge based systems. These problems have been analyzed and an effort have been made to find solutions for the problems related to earlier processes and models for supervised learning and knowledge based system development respectively. Initially this research work concentrates upon investigating past developments in the areas of data mining, supervised learning as well as software and knowledge engineering to find the past developments through literature survey. Then this research work concentrates upon enumerating different processing methods used for supervised learning. These different processes are devised through a general Knowledge Discovery in Databases (KDD) process. From this general process, six specific processes were identified to be widely practiced worldwide. After investigating different processes for supervised learning, limitations of these methods are identified. After identifying problems related to different supervised learning processes, investigations are directed towards finding the solutions for these problems. Search for finding solutions of these problems motivated to develop a new process for supervised learning. This motivation resulted in the form of a new process named as Fuzzy Boundaries of Regression Based Clusters process. Proposed process is based on parallel processing of classification as well as regression algorithms. Proposed process allows pruning the training data without compromising the performance of outcome. It reduces the size of the train set significantly, while analyzing each record of data qualitatively as well as quantitatively through classification and regression algorithms respectively. Proposed process as well as previously known processes are applied on twenty classification algorithms and five regression algorithms over ten datasets gathered from internationally renowned organizations.