Intelligent analysis for multi-level data-driven prediction

Prediction is one of the typical applications in the research fields of machine learning and data mining. Traditional predictive analytics focuses on estimating class membership or numeric value within a specific domain or region, which results in the development of classification models and regression models. However, the restriction on isolated information and the progress of computation technology jointly require the collection and connection of fragmented data for further investigation. Predictive analytics is nowadays expected to be extended its research area to seeking for relationships amongst separated data, which is widely known as link prediction or link analysis. In this thesis, a novel predicting system performing the tasks of instance based missing information estimation, feature variable identification, and variable group pattern recognition has been presented. Specifically, an advanced regression model embedded with both hard and soft clustering techniques is novelly proposed to forecast missing feature values for objects of interest. Also, a link based model creatively employing the concept of connected-triple, and implemented with fuzzy logic, is invented to measure correlations between domain feature variables from various information sources. The resulting link based model has been further utilised as a foundation to construct an adaptative hierarchical knowledge base for describing feature variables under consideration, facilitating both dynamic updating and immediate query. The adaptability and flexibility of the proposed work, together with its remarkable initial performance in sample applications are illustrated by experimental evaluation via datasets collected from both real-world domains and artificial production. The outcomes of comparative studies demonstrate the efficacy of the present work and its great potential for future use. Further suggestions on development and refinement of this research are provided to stimulate inspiration on improving the current work.