Training a Predictive Model with Terrene
With Terrene you have the choice of training either a General Purpose Predictive Model or an Anomaly Detection Model. Predictive models are good for, as the name suggests, predicting future outcomes such as sales projections, inventory needs, or in this case survival on the Titanic. Anomaly detection models are good for recognizing discrepancies in data. For example, recognizing fraudulent credit card transactions or recognizing odd computer activity indicative of a cyber attack.
If you are using your own data and need an anomaly detection model click here. Otherwise continue onward to learn how to train a predictive model of the Titanic data.
Now that you have uploaded the Titanic data you get to choose what variables you are going to compare. At the bottom of the page, you can see a statistical breakdown of the variables in the dataset which can help inform your selection. In this example, I am going to select Age, Sex, SibSp (number of siblings or spouses on board) and Parch (number of parents on board). You can choose the same variables or make your own selections to see how different variables affect survival. I will recommend not using Name, PassengerID, or ticket because they are identifiers and they have no impact on the outcome. Also, make sure you do not select any variables that will become outputs (in this case survived).
Now that you have chosen your input you can select your output, or what you are trying to predict. In the case of the Titanic, we will be trying to predict survived.
Once you have selected your variables you can click "Finish" and Terrene will begin training a deep learning neural network around your data and you can start making predictions!