Let us assume that, we have a company called "Splash", and you are a newly joined Machine Learning Engineer at our company, held at 30 LPA CTC. So, the below thing describes our company and the task in hand.
Currently we at Splash, help retailers with pricing intelligence and business monitoring.
One of our main challenges comprise of monitoring and analyzing our client's business metrics in real time for instant detection of the incidents that may impact their revenue.
One subpart of this challenge is Anomaly Detection which generates alerts on our client's business metrics.
The Problem Statement presented below highlights the problem which we are tackling right now
Column Description
The description of the column are as follows:
timestamp [ float ] : is provided as a Unix epoch in seconds.
value [ int ] : is a real value measurement of some metric at the timestamp.
is_anomaly [ boolean ] : is a boolean value which is True if the corresponding value is
identified as an anomaly.
predicted [ float ] : is a real value prediction coming from a black box forecasting
model for that timestamp. This black box forecasting model is assumed to be aware of
only the true data distribution.
Submission Format
The CSV format should contain two column, timestamp, is_anomaly score
Scoring Criteria
The scoring will be determined with the help of the Mean-F-Score, because the prediction of the anomaly is the important part, merely just making all the values to be False, would result in maybe 95% accuracy, which is not the desired thing for the client, because he is concerned more about the anomaly rather than the non anomaly data.