Iterative Bayesian fuzzy clustering toward flexible icon-based assistive software for the disabled

A novel fuzzy clustering technique, called iterative Bayesian fuzzy clustering (IBFC), is presented and applied for grouping and recommendation of icons associated with assistive software meant for the physically disabled. The algorithm incorporates a modified fuzzy competitive learning structure with a Bayesian decision rule. In order to ignore unintended behavior of the user, a Bayesian minimum risk classification rule with two loss coefficients is built into the algorithm. This provides a rational basis for outlier detection in noisy data. In addition, we show that the inclusion of a unique control parameter of IBFC allows for establishment of a strong relationship between learning region and cluster congestion. This interpretation leads to an agglomerative iterative Bayesian fuzzy clustering (AIBFC) framework capable of clustering data of complex structure. The proposed AIBFC framework is applied to design a flexible interface for the icon-based assistive software for the disabled. The latter is utilized in grouping and recommendation of icons. Additionally, the proposed algorithm is shown to outperform several well-known methods for both IRIS and Wisconsin benchmark data sets. Finally, it is shown, using a questionnaire survey of real end-users, that the software designed using AIBFC framework meets users' needs.