The demand for accurate Multi-Output Spatio-temporal Forecasting is rising in areas like public safety, urban mobility, and climate variability. Traditional methods struggle with model calibration and data integration. This paper presents a methodological guideline for creating forecasting pipelines that handle multi-output forecasting complexities. Using a uniform methodology tested on three diverse datasets, the framework combines genetic algorithms and advanced models to optimize forecasting. Our evaluation shows significant performance improvements, with better adaptability to urban and rural datasets, aiding decision-making in spatio-temporal analysis. The framework achieved a 20% average improvement in the R-2 metric across all datasets, outperforming benchmark models.