Evaluate Model

Training the Machine Learning Model

Now that we have collected our dataset, it's time to train our model.

In this tool, all we need to do is go to the 'Train Model' tab and click on the 'Train Model' button. You will see the model's progress on the page. The graph on the left shows the accuracy, it should converge to 1, and the graph on the left shows the loss or the amount of error, it should converge to 0.

We can also evaluate our model by looking at how well it performed on the training dataset. A confusion matrix is a plot that compares the ground truth labels to what the model predicts. The confusion matrix seen here suggests that the model learned to distinguish many of the classes well, but struggled with the "still" class.

In the context of the problem we are trying to solve, this gives us two important pieces of information about our model. First, that there are major accuracy issues with an important class. Stakeholders value fairly accurate predictions, and a model that incorrectly classifies "stillness" as "taps" half the time would be very frustrating.

Second, there may be issues with using this model with a diverse group. A common problem with machine learning models is overfitting. This is when a model learns its training dataset too well and cannot generalize to new data points. We should make sure our model is not overfitting by testing the model on data that is different from what we used to train our model. Importantly, we should make sure the model works equally well for different subgroups of our stakeholders.