Application of ANFIS and LLNF models to automobile fuel consumption prediction: A comparative study

Dividing the input space into smaller parts, fitting local simple models in each part and then combing the outputs of local models to obtain the overall output is one efficient approach in applications like modeling, prediction, etc when the input-output relationship is complex. Adaptive network based fuzzy inference system (ANFIS) and Locally Linear Neuro-Fuzzy models are among the most successful models for such applications. On the other hand selection of the most appropriate input variables from the available candidates is crucial for successful modeling, prediction, etc. In this paper after investigating the input selection issue for choosing the best inputs, the LLNF model with an efficient learning algorithm (LOLIMOT) is utilized for a typical prediction problem, the automobile fuel consumption prediction, and the results are compared to those of ANFIS network. Simulation results show that the LLNF model with LOLIMOT learning algorithm has surpassed the ANFIS network for this application.