ML Model - SARIMAX

[LB_Team]

Hi all,

I'm working on time series forecasting using SARIMAX in Python, and I encountered a confusing result:
My dataset is count of tests taken on daily from the year 2022 jan to 2024 dec 

• When I use this data and forecasted for next 12 months using SARIMAX gave me accurate forecasts.

• But when I use the same data and tried to forecast for the next 7 days instead of 12 months, the predictions are often inaccurate sometimes showing a steep decline or not capturing the correct trend at all.

Where am I going wrong? Please guide me.

I've attached the link to the code :
https://drive.google.com/file/d/1fhOf-bY0oZTYyq9wifXRDEaHa659-pxp/view?usp=sharing

[v-sdhruv]

Hi @LB_Team ,

If your data shows seasonal patterns and you want to forecast for 7 days, then SARIMAX is sensitive. Your model probably learned seasonal trends well over long periods, but struggles to generalize for short-term prediction due to overfitting to long cycles or reacting too much to end-point anomalies.

Check residuals near the end of your training window.

If residuals are high, the model is already misfitting — so short-term forecasts will diverge quickly.

Tailor your model specifically for weekly forecasts-->s=7

Use smaller seasonal_order and possibly a narrower training window-- SARIMAX(..., seasonal_order=(P,D,Q,7))

I hope this helps!
If the response has addressed your query, please accept it as a solution so other members can easily find it.
Thank You

[LB_Team]

But my data shows seasonality patterns also will ARIMA be suitable for my usecase to forecast for next 7 days?

[v-sdhruv]

Hi @LB_Team ,

SARIMAX captures seasonal components, trends, and exogenous variables. If your model includes seasonal components (e.g., weekly or yearly), then:

Long-term forecasts (like 12 months) are more influenced by seasonal patterns it has learned.

Short-term forecasts (like 7 days) might be more sensitive to noise, recent variability, or model overfitting to short-term anomalies.

Improper seasonal or non-seasonal orders can have unexpected short-term behavior. For example:

1.Over-differencing can cause forecasts to trend downward.

2.Misidentified seasonal periods  can distort short-term forecasts

Try using ARIMA model:
Refer-
https://skforecast.org/0.10.0/user_guides/forecasting-sarimax-arima

Hope this helps!
If the response has addressed your query, please accept it as a solution so other members can easily find it.
Thank You