Fast and Efficient Parallel Execution of SARIMA Prediction Model

Mathematical models for predicting values in time series are powerfultools for the process of knowledge discovery and decision making in several areas. However, the choice of the predictive model and its configuration are not trivial tasks, requiring a long processing time to obtain the results due to the high complexity of the models and the uncertainty of the value of the best parameters. Calculations performed by these approaches use sampling from the dataset, which can present discrepancies and variations that can directly impact the final result. Therefore, this work presents a new approach based on the SARIMA model for the prediction of values in time series. The proposal aims at predictive calculation from multiple executions of SARIMA in parallel, configured with predefined order and seasonal order parameters and applied to values already known in a time series. Thus, from the results obtained in past observations, it is possible to determine the percentage of precision that each parameter obtained, and, in this way, to determine the parameters that are more likely to obtain more accurate values in future observations, thus, eliminating the need to use specific algorithms to estimate them. The proposed approach is capable of achieving results with greater precision and performance compared to the traditional SARIMA execution, achieving results with greater assertiveness, reaching up to 10.77% of better accuracy and with better processing times, without the need for validation and parameter adjustments required by the settings obtained by functions, such as ACF and PACF.