Vocavio has just concluded a project with newly founded data science company, B&P Consulting based in Ontario, Canada. The project brief was initially focused on mapping OBs (observable behaviours) annotated by flight instructors to signal data generated by Vocavio’s voice analysis. This sprint project quickly became about the sample size available, what models might support such analysis and what else can the data share with us?
In summary, here’s what the data did tell, and some insights might just surprise you.
- Top instructors observed 82% of what the voice analytics sensor observed in relation to workload during a 20 min LOFT exercise. Average instructors observed 40% and lower yield was at 16%.
- Lag between instructors observing and inputting on e-grading is to be expected, averaging at 46seconds. The extreme lag was at 160 seconds between behaviours observed and data annotated into the e-grading solution. This lag impacts on model fit by reducing the validity of the instructor signal tied to more ambiguous workload/stress events.
Read full article here: