Putriaji Hendikawati, Subanar, Abdurakhman, Tarno
In reality, most time series observations take the form of multivariate data that are influenced by many factors. In real-world modeling problems, too many inputs can increase calculation complexity due to the many parameters that must be estimated and resulting in reduced accuracy. This study uses the Adaptive Neuro-Fuzzy Inference System (ANFIS) method to apply various data preprocessing techniques, such as regression, Autoregressive Integrated Moving Average (ARIMA), and Autoregressive Integrated Moving Average with Exogenous Variable (ARIMAX), for the determination of potential input variables for time-series data subject to the calendar effect. The hotel room occupancy rate in the Special Region of Yogyakarta (DIY), which is influenced by the calendar effect, is predicted with this method. Preprocessing and correct sampling from, input data can have an impact on the prediction results. In general, data preprocessing improves efficiency. The empirical study shows that ANFIS preprocessing with the ARIMAX model provides the best results. This model obtained the smallest root mean square error (RMSE) for training and testing under the ANFIS model, i.e., 26,025.779 and 67,468,167, respectively. This empirical study shows that the preprocessing data that has been corrected according to calendar variations will positively impact the prediction performance. For ANFIS architecture, it can be considered to use triangular and gaussian membership functions with a minimal number of clusters and the grid-partitioning clustering method. © 2021. All Rights Reserved.
Mathematics Department, Universitas Gadjah Mada, Yogyakarta, 55281, Indonesia; Mathematics Department, Universitas Negeri Semarang, Semarang, 50229, Indonesia; Statistics Department, Universitas Diponegoro, Semarang, Indonesia
