Pilot Intelligent Training Systems
DTW-MDS embedding technique.
This paper presents a new technique for representing time series data that combines dynamic time scaling (DTW) and multi-dimensional scaling (MDS) to create a distance embedding for modeling the student performance of flight tasks in simulation.
Our research questions are:
• RQ1: is it valuable to preserve the temporal ordering of the data for student modeling using a time series or to vectorize the data using the distribution features?
• RQ2: does the same representation perform well across multiple machine learning tasks?
We hypothesize that disparate student modeling problems are not likely to benefit from the same distance embedding; hence the data preprocessing pipeline must be configured for the research problem.
-
Experiment Results:
Comparison with previous works on flight success prediction. 
Performance of different components of MDS.