AI-driven help prediction: classify user behavior with CoreML, iOS

The purpose of this article is to collect analytics data and identify users who require assistance, categorizing their needs accordingly.

Disclaimer: This is a demo application, and it does not use any user data.

Before we begin our experiment, there are certain terms that we should establish.

• Do not collect users’ data;
• Hence, it is not possible to send any data to the server, and only a solution that can be performed on the device is viable.;

Let’s determine our data collection. Each data point will be an “Event” containing:

• Screen name
• Duration time on a specific screen

Event recorded by our custom AnalyticService. At some point, every user behavior should be labeled to classify which type of help users need. The list of different types of support we provide can be easily extended.

I’ve created a demo app to reach the following goals:

• Generate analytics data
• Try out the prediction model
• Proof of concept

Playground project