In my economic forecasting project using time series data, I'm getting inconsistent results when applying ARIMA models, especially with seasonal adjustments. What are the best practices for model selection and validation in such cases?

I'm working on a university assignment where I need to forecast quarterly GDP growth for a small economy. I've collected data from the past 20 years and tried fitting ARIMA models in Python with statsmodels. However, when I add seasonal components, the AIC and BIC values fluctuate wildly, and the forecasts don't align with historical trends. I've already checked for stationarity using ADF tests and differenced the data appropriately, but I'm unsure how to proceed with model diagnostics or if I should consider alternative methods like SARIMA or machine learning approaches.