SeriesFit2LinesEstimateUsage.kql

//using series_fit_2lines() - useful when there is a sudden change in the data
let timeRange = 90d;
let projectionDays = 30;
Usage
| where TimeGenerated > ago(timeRange)
| make-series logConsumption = sum(Quantity) default = 0
on TimeGenerated in range (ago(timeRange) , ago(1d), 1d)
| extend (RSquare,SplitIDx,Variance, RVariance, LineFit, right_rsquare,right_slope,right_interception,right_variance,right_rvariance,left_rsquare,left_slope,left_variance,left_rvariance) = series_fit_2lines(logConsumption)
| extend numberOfDaysUsed = timeRange/1d
| render timechart

//using series_fit_2lines() - useful when there is a sudden change in the data
let timeRange = 90d;
let projectionDays = 30;
Usage
| where TimeGenerated > ago(timeRange)
| make-series logConsumption = sum(Quantity) default = 0
on TimeGenerated in range (ago(timeRange) , ago(1d), 1d)
//apply two segments linear regression to the logConsumption array to determine index of split, right slope, right nterception, LineFit
| extend (RSquare,SplitIDx,Variance, RVariance, LineFit, right_rsquare,right_slope,right_interception,right_variance,right_rvariance,left_rsquare,left_slope,left_variance,left_rvariance) = series_fit_2lines(logConsumption)
//convert time range to scalar value
| extend numberOfDaysUsed = timeRange/1d
//using index of split point (days from the begining of the time series) calculate estimated log usage in 30 days from the current date
| extend estUsage = right_slope*(numberOfDaysUsed - SplitIDx + projectionDays) + right_interception
| project estUsage