From ARIMA to Chaos-Enhanced LSTM: Methodological Progression of Philippine Forecasting Models Across Energy, Agriculture, Finance, and Health

Forecasting research in the Philippines has evolved from classical time-series analysis toward comparative machine learning and selected hybrid nonlinear models. Across energy, agriculture, health, and finance, this progression reflects changing data complexity, forecasting goals, and temporal behavior rather than simple replacement of older methods. Early studies on electricity consumption, provincial demand, and coal production relied on ARIMA-type models because the series were aggregate, limited, and well suited to interpretable linear frameworks. Later work on crop production, monthly peak demand, and disease incidence adopted comparative designs that evaluated seasonal statistical models alongside neural and machine-learning approaches, showing added value when seasonality was less stable and nonlinear dependence was more evident. The progression culminated in chaos-enhanced LSTM modeling for the Philippine stock exchange index, where nonlinear state reconstruction became methodologically defensible. Overall, ARIMA remains sufficient for largely linear seasonal signals, whereas machine learning and hybrid methods are justified for more complex dynamics.