You are here:
Imbalanced Distribution Alert
Imbalanced distribution indicates a disproportionate ratio of observations in each class of the training data.
Actions to Consider
Training a model on an imbalanced dataset can result in a less reliable model with biased predictions. When you create a model version, Einstein:
- increases the frequency in the minority class by upsampling the data.
- correct values that reflect an extraordinary, non-recurring, or infrequent event.
Detection Methodology
Model Builder automatically alerts you when it detects an imbalanced distribution.
Example
An e-commerce company wants to predict whether its customers will cancel their subscription or churn. To achieve this, the company builds a binary classification model with these input variables.
- customer ID
- monthly cost
- monthly login frequency
- subscription type
- tenure
- churn
After model training, an alert displays because the variable “churn” has a disproportional class distribution, with 95% of customers labeled as active and only 5% as churned. This imbalance can cause the model to favor predicting “active" customers, reducing its ability to accurately identify the ones that churned (high false negatives). To resolve the issue, here are some actions to consider:
- Follow recommendations provided by the alert to balance the dataset.
- Retrain the model with an updated dataset.
