Abstract
<jats:p>Accurate time series forecasting plays a key role in decision support systems used in economics, industry, and engineering. This article discusses a time series forecasting model based on an adaptive neuro-fuzzy inference system (ANFIS), designed for modeling nonlinear relationships with limited historical data. Particular attention is paid to the model's implementation, taking into account compatibility with the MATLAB environment, which expands its practical application in the context of limited computing resources. To assess forecast quality, the mean absolute error (MAE) and mean absolute percentage error (MAPE) metrics were used, allowing for an objective assessment of the accuracy and robustness of the resulting forecasts. Experimental results confirm that the proposed ANFIS model provides reliable short-term forecasting accuracy and demonstrates stable convergence during training. A significant advantage of the method is the interpretability of the model, ensured by the use of fuzzy rules and membership functions, which allows for analyzing the influence of past time series values on forecast formation. The obtained results demonstrate the feasibility of using neuro-fuzzy systems in time series forecasting, particularly in situations where a combination of accuracy, robustness, and explainability is required. Further research areas include the development of hybrid models, optimization of ANFIS parameters, and comparative analysis with modern deep learning methods. Keywords: Time series forecasting, adaptive neuro-fuzzy inference system (ANFIS), mean absolute error (MAE), mean absolute percentage error (MAPE), MATLAB.</jats:p>