Integrated knowledge-based modeling and its application for classification problems

Knowledge discovery from data directly can hardly avoid the fact that it is biased towards the collected experimental data, whereas, expert systems are always baffled with the manual knowledge acquisition bottleneck. So it is believable that integrating the knowledge embedded in data and those possessed by experts can lead to a superior modeling approach. Aiming at the classification problems, a novel integrated knowledge-based modeling methodology, oriented by experts and driven by data, is proposed. It starts from experts identifying modeling parameters, and then the input space is partitioned followed by fuzzification. Afterwards, single rules are generated and then aggregated to form a rule base, on which a fuzzy inference mechanism is proposed. The experts are allowed to make necessary changes on the rule base to improve the model accuracy. A real-world application, welding fault diagnosis, is presented to demonstrate the effectiveness of the methodology.