We now have a model in place that gives us relatively good accuracy. As a ballpark figure, we can say that we achieved an overall 75% accuracy with a TPR of 70% and TNR of 75% (scope exists to improve this further).

How does it add up to the use case's revenue story? With our model in place, we can say that we will correctly predict a power outage 7 out of 10 times. So we have saved the losses that happen because of power outages by 70%. Now, we also incorrectly predicted a power outage when it wasn't, that is, approximately 2.5 out of 10 times. Let's say that there was a cost associated with stocking diesel when it was predicted that the next day will be having power outages; this cost will tax the losses saved from the correct prediction (penalty).

The overall FPR is low and also the cost associated with stocking diesel for power outages is generally much lower than the losses due to unplanned power outages. Therefore, we are still in a good position to add value to...