Talk Data to Me: Accurately Predicting District Absences

Last fall, the Frontline Research and Learning Institute analyzed over years of absence data and reported high-level predictability of K-12 absence trends. That analysis, further outlined here, lead to the development of a district-level absence prediction model by Frontline’s Human Capital Analytics team and districts across the county utilized the model’s results during the 23-24 school year.   

So, how did the model perform? How accurate was it at predicting a district’s absences? Let’s take a look!  

Absence Prediction Data  

The Absence Prediction model inside of Human Capital Analytics provides a district with a predicted number of substitute-required absences that they can expect to occur in their district on any given day. The model, which is trained using each district’s unique historical absence data, also provides a measure of accuracy to the district so they can be confident in its predictions.  

The accuracy measure takes the daily difference between the district’s prediction and the real number of absences that occurred each day, and divides that average difference by the average number of real daily absences that have occurred in that district. That result gets subtract from the number 1 to produce a percentage measure where the larger the percentage, the more accurate the model has been.   

For example, let’s say the average daily difference between the prediction and reality is 4 absences, and the district has an average daily absence total of 40 absences.  

Across the sample of K-12 districts utilizing Human Capital analytics during the 23-24 school year, their models were highly accurate at predicting district-level absence totals.  Check out the table below!  

Accuracy Range	% of Districts
90-100%	68%
80-89%	19%
70-79%	10%
Less Than 70%	3%

The takeaway: 

• 67% of districts enjoyed daily predictions that were greater than 90% accurate. 

• A whopping 87% of districts benefited from daily predictions that were at least 80% accurate.   

Districts were able to use their predictions to proactively allocate substitute resources, optimize positioning of building substitutes, set up incentive programs to encourage substitutes to fill absences on days where high volumes of absences were predicted, and intelligently plan professional development that would minimize the creation of absences on certain days.   

Want Your Own Predictions?  

With Human Capital Analytics, you can access machine learning predictions for the entire 23-24 school year. This tool will empower district leaders to analyze which days they can reasonably expect more or less sub-required absences and see how their absence trends rise and fall throughout the year.