Limits of machine learning – What the Forecast app cannot do
Although the Forecast App based on powerful LSTM models, it is – like any machine learning method – only as good as the data it works with. This chapter highlights typical pitfalls and misconceptions that can lead to poor or misleading results when using the app.
1. No prediction without a pattern
Problem: If your data does not contain a recognizable pattern, the model cannot find one either.
Example: If a target attribute fluctuates purely at random or contains no relevant correlation to the input values, then the Forecast App can only estimate average values, but cannot make reliable predictions.
Recognizable by:
Flat prediction curves
Little model improvement after retraining
Very high or very low confidence interval in the output
Solution:
Check in advance in Analytics & Reports, whether recognizable trends or patterns are present.
Avoid binary, random, or irregular signals without explainable dependencies.
2. Extreme values and outliers disrupt the model
Problem: A single outlier (e.g. due to a faulty measurement) can strongly distort the model structure and prediction behavior.
Example: If all values are in the range of 0.01, but one data point suddenly is 10,000,000, the model will artificially raise future predictions to account for possible "similar" outliers.
Solution:
Clean historical data before training (e.g. by filtering or manually removing).
Use logarithmic or normalized values if necessary.
Use the Eliona Calculatorto derive differentiated or smoothed values.
3. New value ranges lead to inaccurate forecasts
Problem: If values suddenly appear after training that the model has never seen before, it cannot learn any meaningful response to them.
Example: A sensor delivers values between 10–50 during training. After training, new values in the range 100–200 appear. The model does not "know" this range and behaves unpredictably.
Solution:
Instead of predicting absolute values, calculate relative changes or differences.
If the data behavior changes: retrain the model.
4. Models require sufficient data
Problem: If there is too little data, no viable model can be created.
Recommendation:
At least several hundred data points should be available.
Ideal: histories with seasonal or periodic fluctuations over several cycles.
The system automatically checks the data length and starts training only when enough measurements are available.
6. Frequent structural breaks in the data trend
Problem: Sudden changes (e.g. system conversions, changes in measurement methods) lead to structural breaks that the model cannot explain.
Solution:
Segment the dataset before training.
Avoid mixing different data sources or measurement methods in a forecast.
Conclusion
The Forecast App is a powerful tool – but only with sensibly prepared and structured data. It does not recognize meaning, only statistical patterns.
The model cannot:
Independently detect faulty data
"Understand" or interpret decisions
Deal sensibly with completely unknown data ranges
Make predictions when no explainable patterns exist
But the model can:
Recognize regularities
Derive reliable trends from historical patterns
Provide forecasts based on consistent, cleaned data
Good forecasting does not start with the model – it starts with the data.
Was this helpful?