Metaheuristic-driven automatic ARIMA order selection using spider monkey optimization for agricultural production forecasting.

The Autoregressive Integrated Moving Average (ARIMA) model requires appropriate selection of autoregressive (p), differencing (d), and moving average (q) orders for accurate forecasting. Conventional order selection methods, such as grid search and stepwise information-criterion approaches, can be computationally expensive and prone to suboptimal solutions. This study proposes ARIMA-Spider Monkey Optimization (ARIMA-SMO), an automated framework that employs Spider Monkey Optimization to identify optimal ARIMA orders. In the proposed approach, each spider monkey represents a candidate ARIMA (p,d,q) configuration, and model fitness is evaluated using a combination of statistical goodness-of-fit and forecasting accuracy. The local and global leader mechanisms, together with adaptive subgroup restructuring, balance exploration and exploitation of the search space. The method was evaluated using agricultural production time series from India and compared with Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Bayesian Optimization (BO)-based ARIMA models. Across all six-production series, ARIMA-SMO consistently achieved the lowest forecasting errors while requiring less computational time than the competing approaches. For the total food grains series, ARIMA-SMO reduced the test RMSE to 35.75 compared with 43.96 (GA), 39.84 (PSO), and 37.12 (BO). The results demonstrate that ARIMA-SMO is an effective and computationally efficient framework for automatic ARIMA order selection and time-series forecasting.