Once you successfully connect your data to your Obviously AI account, you can start building models with it. When a model is built a report with a name such as Model ABCD will be automatically generated. This name is random and autogenerated by the platform but you can edit it anytime just by typing directly on the name. Your model will consist of the following tabs:
This article details the Summary tab of your trained model - the different sections and the meaning of all the metric values for both AutoML and Time Series.
For a classification task we have the following details on the Summary tab of the model. On the left you’ll see all the models that you’ve built so far on your platform, however you can hide this menu and just concentrate on your current model (as shown below). Also note that we have changed the name of the model to Model Churn.
The Model Summary section highlights the prediction accuracy of the model, the best performing model chosen by the platform (since we run multiple models in parallel), the top 3 feature columns with the most impact on the prediction column Churn and the prediction quality where we highlight each class count, F1 score and their quality depending on the F1 score value.
Right below the Model Summary section is the Drivers section. Here you can check the %impact each feature column has on the prediction column, Churn = No. You can toggle between the classes here. We show the top 10 drivers, and add the %impacts of all the other drivers and display that at the end. Moreover, when you click on any one driver, on the right you’ll find how different ranges of values within the driver impact the churn.
For example, the top driver here is TotalCharges with ~28% impact. On the right you see that when the value of TotalCharges is between “6947 - 7382” it increases the non churn of customers by 6.6%.
Finally, the bottom section showcases the ideal personas depending on the number of classes in your data. Here we have only 2 classes. Each persona highlights the best possible combination of feature values for Churn=No (Persona #1) and Churn=Yes (Persona #2).
For regression analysis the model Summary tab is similar to classification, the main difference being here we are displaying the performance metric R2 score of the platform chosen best model, since we are now predicting real numbers in case of regression (instead of classes)
Under the Model Summary you’ll see the name and R2 score of your trained model. Additionally, you’ll also see the top 3 features in your dataset that affect your prediction column, here fare_amount. Finally, the prediction quality that reflects the mean absolute error (the lesser the value the better)
Next we move on to the Drivers section. Similar to what we mentioned under classification, the drivers are the features ranked in order of their % impact on the prediction column. We show the top 10 drivers and add the % impact of all the other drivers and display them at the end. When you click on any driver, on the right you’ll find the distribution of that driver’s different ranges of values and how they affect the prediction column (displayed as % values, green for positive and red for negative)
Finally, we check the ideal personas for this model. Here, for numbers we have the highest fare_amount and lowest fare_amount. As seen on the screenshot here, the highest is 57.65 and the lowest is 3.37