Predict service time from features such as service branch ID, technician ID, and location of service call.
We use features such as machine id, call priority level, hour of day, day, month, year, state, zip code, service branch code, customer type code, and representative id to predict a range for time required to complete the service call. We took a multi-class approach as we are interested in predicting a broad range—a 15-20 minute difference is acceptable—to get an approximate idea of service call duration. The duration ranges are: 0–179, 180–359, 360–719, 720–1439, 1440–2999, 3000+ Moreover, among other things, the range of service call duration (0 to 100000 hours) was very large, and exact duration was not required. Therefore, we did not take a regression approach.