Identifying Device Abuse Patterns

To minimise escalating warranty claims and costs for devices damaged in use, a leading industrial handheld mobile device company needed to better understand device usage patterns that drove device damage. This involved detecting types of device uses from device movement patterns and separating proper device use from device abuse, thereby allowing a data-based assessment to be made about applicability of Device Warranties.

Using Machine Learning to detect Device Accidental Fall vs Abuse

To detect device movements, an on-device custom-Android agent was implemented to capture and aggregate device accelerometer data on an AWS cloud data store. To identify event-signatures of damage-drivers, Machine Learning algorithms were then trained on this accelerometer data. Event-signature were identified for device drop, impact, throws, slam (on a table or wall) and accidental drops while also identifying false positives for running with the device in the pocket or throwing the device followed by catching it. 

Using these, Device abuse use cases could be separated out and classified with  ~90%  accuracyThis also helped to  provide quicker maintenance and improve device  ruggedization measures.