Classical regression analysis assumes that the data belong to a single class. However, in many application studies, data are typically gathered from various mixed classes without prior knowledge of each class’s membership. Switching regression analysis, which creates suitable sub-models for each different class, is used to model such data. As the quantity of independent variables and classes within the data increases, the number of sub-models to be generated simultaneously in switching regression also increases. In such scenarios, it is recommended to utilize the Adaptive Network-based Fuzzy Inference System (ANFIS), which proves to be an effective tool for addressing mixed problems and systems.  Achieving successful training is essential for obtaining effective results with ANFIS. This can be achieved by determining its structure’s premise and consequence parameters with practical optimization algorithms. This study focuses on training ANFIS with Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). The effectiveness of the GA-trained ANFIS and PSO-trained ANFIS are tested with a numerical example where the dependent variable has an outlier. The results show that the proposed methods, especially PSO-trained ANFIS, provide accurate predictions and are not affected by the outlier in the dependent variable.