Data 360: Addressing Model Training Failures Due to Lack of Variation in Outcome Variables

When training a predictive model, it is crucial to have variation in the outcome variable. If there is no variation, meaning the outcome field (e.g., boolean field) contains only one type of value (e.g., all false values), the model cannot learn and make accurate predictions. For effective model training, the chosen field must have more than 1% different values.

For example, if you have 10,000 sample records being used to train a binary classification model, make sure that at least 101 of those records have true values, while the remaining records have false values, or vice versa.

If the outcome field lacks variation, you will encounter an error message when attempting to run the Einstein model: "There isn't enough variation in the outcome variable. Add more data with variation or choose another outcome variable."

To resolve this issue, ensure that your dataset includes a sufficient amount of different values in the outcome variable. This variation is essential for the model to understand and predict different outcomes accurately.