Blog Digital Twin

How to check the model due to an error

While adding new data (more specifically, processing new 220 countries and 1700 municipalities) to our service, we’ve made a mistake, due to which the model received data only until 2008th year. Meaning, the “new” values actually lost their relevance 15 years ago. But once we figured that out, we haven’t instantly corrected that – instead we wondered what the model will show if we compare its forecast with actual data. Since our model hasn’t “seen” any data for the present, our present became its future – which was the perfect opportunity for evaluating its forecasting abilities.

We took Germany and its population as an example.
The model anticipated growth in spite of “recent” (in models’ eyes) descending trend, and it turns out that for 2021th — 13 years horizon — our model predicted 83.4 million people for population. In fact, the actual Germany population in 2021 is 83.2 according to the UN. That is, on a horizon of 13 years, the forecast error is 0.2% for absolute value. And this is how we once again checked our models.

Link to the service: