Hi, I'm experimenting with hwpredict failure detection in business activity data (detecting anomalies in mass scale business processes) and have a problem with false positives at night.<div>Since the activity at night is very small there's no way to accurately predict it - usually the statistics are at zero+epsilon, but even a small number of events (like few customers sending complaints at midnight) generates a spike that's outside the predicted deviation. Also, the value of zero is quite often outside the deviation (because the average during the night is slightly greater than zero, but individual spikes occur randomly and cannot be predicted).</div>
<div>So i think the best solution would be to disable the failure detection at night time (for example between 10pm and 6am). Can you suggest an elegant way of doing that?</div><div>Best regards</div><div>RG</div>