Hello everyone,

I’m currently working on a time-series forecasting model for product demand, and I’ve encountered some challenges related to seasonality and unexpected trend shifts. While analyzing the data, I noticed that certain seasonal patterns seem to change over time—some peaks shift slightly, and in a few cases, long-term trends reverse abruptly, likely due to external events (e.g., promotions, market disruptions).

I’d appreciate advice on the following points:

Seasonality detection: What are the most reliable ways to detect and validate seasonal patterns in a time-series dataset, especially when seasonality may not be stable over time?

Handling trend shifts: How do you adjust your model when a significant change in trend occurs? Is it better to retrain the model more frequently or to include exogenous variables that may explain the change?

Model robustness: Among forecasting methods like SARIMA, Prophet, or transformer-based models, which have you found most adaptable to evolving patterns and structural breaks?

I’m particularly interested in hearing about real-world strategies you’ve used to maintain forecasting accuracy when seasonality and trend behaviors change over time.

Looking forward to your insights!

Best, Among Us Online