Cluster driven approach for pattern identification and understanding

Conventional practice of data mining is still highly dependant on human intervention. In particular, we argue that conventional data mining approaches do not initiate pattern discovery because humans are still greatly involved in justifying the outcomes and the pre-coded patterns that appear in their minds. This means that possible “pattern” must be “represented” in the human minds before the deployment of data mining algorithms that search for patterns that humans want to find. Thus, this has created the dilemma - Do data mining techniques actually find hidden patterns or are they finding patterns which are pre-coded by humans? Pattern observation and detection are what humans do on daily basis because we have the unique ability to perceive and detect both simple and complex patterns naturally. In fact, we are also capable of perceiving things from different perspectives. Biologically human brain processes information in both unimodal and multimodal approaches and progressively information is abstracted and fused seamlessly. Similarly, the accumulation of human experience and knowledge essentially originates from multiple sources of inputs and information. These multimodal inputs and information are then processed and fused to obtain holistic understanding of a problem. Subsequently, such processes that naturally occur in our brain could be used as an inspiration for us to develop a methodology for data explorations as data that we collect also consists of multiple modalities. Thus, the main aim of this n thesis is to investigate and propose a novel conceptual model that explores and identifies patterns based on the biological brain structure and cognitive perspective of the human mind. Specifically, we propose that patterns could be explored and identified at different levels of granularity, different types of hierarchies and different types of modalities. A structurally adaptive neural network based implementation strategy is deployed to implement the conceptual model. The conceptual model is applied in a real application domain where the functionalities of the proposed model are demonstrated.