Simulation of an equivalent reduced order system from large, imprecise, and uncertain data system using multistage multivariate analysis and neuro fuzzy approach

System reduction by multivariate analysis has been a topic of the research work for the system theorists when large systems are often unwieldy to handle. Fuzzy logic systems deal with input/output space by generating linguistic rules. Even though high-speed computational power is available today, it is very difficult to utilize large input/output systems since the system generates a very large number of linguistic rules, which are difficult to handle, due to the limitations of time and memories.; The purpose of this dissertation is to develop a reduced order system from the original large system, so that the behavior of large fuzzy system and reduced order fuzzy system is approximately same. In this dissertation, a number of algorithms, called Multistage Multivariate Analysis (MMA), are proposed, which determine an equivalent reduced order fuzzy system without losing any significant meaning. To verify MMA algorithms, three different cases of system reduction are considered: data validation, data classification, and data visualization. For data validation, the combination of factor analysis and clustering analysis or principal component analysis and clustering analysis is utilized. In addition, the behavior of the original fuzzy system and the reduced order fuzzy system is compared and evaluated by the index called “Equally Weighted Index (EWI).” For data classification, the technique of principal component analysis and Hierarchical clustering is employed to the search time of target detection. For data visualization, the modified clustering technique is applied to the robot path planning data.; It is hoped that this research will open a way to develop new and challenging algorithms so as to handle large fuzzy systems for imprecise and uncertain data.