Human centered Neuro-Fuzzy modeling

Human Centered Computing, a recently developed research field has many connotations, one of which is the field of developing easily interpretable, highly transparent and concise modeling approaches. The objective of such an endeavor is to develop technology with high cognitive appeal. Neuro-Fuzzy modeling can incorporate knowledge from the data as well as heuristic knowledge in approximating the system and hence are ideal for human centered computing. However, the training process often results in the degradation of this knowledge and thus the system is not easily interpretable. In this research we demonstrate this phenomenon using a case study and also develop evaluation criteria that can be used to detect such a phenomenon. The evaluation criteria are based on Kullback-Leibler distance which is also used for evaluating the networks in the form of Akaike Information Criterion (AIC).