Loss-of-control events during the approach-to-landing phase of flight account for a large share of fatalities in general aviation. During this critical transition towards the runway it is essential that an aircraft is stabilized. Pilot discretion and judgment is used to determine if an aircraft is suited to either land or go-around, based on an assessment of approach conditions. Many landing incidents and accidents could be prevented with improved go-around decisions. The purpose of this research is to investigate the utility of neural networks in modeling those decisions using historic aircraft flight data. Data collected from nearly 2,000 hours of training flights is used to create a snapshot of an aircraft’s flight parameters at 200’ above ground level on approach. Each approach is then categorized as a landing event or go-around; using this data set a neural network is trained to predict approach outcomes. The network is then tested with an unfamiliar data set. Low error rates with testing data indicate the success of the network in predicting go-around events.
"FDM Machine Learning: An investigation into the utility of neural networks as a predictive analytic tool for go around decision making,"
Journal of Applied Sciences and Arts: Vol. 1
, Article 3.
Available at: http://opensiuc.lib.siu.edu/jasa/vol1/iss3/3