Comparison of Hidden Markov Models and the FAST Algorithm for Feature-Aware Knowledge Tracing

In many Indian rural schools, individual students do not receive adequate attention due to the high student–teacher ratio. It is anonerous task for teachers to assess their students’ knowledge levels, and identify their deficient areas of learning. An Intelligent Tutoring System (ITS) enables the teacher to create a report on topics, which students need to study in more detail. Knowledge tracing approaches are a good option for the generation of such reports. For the current paper, we analyze first-grade students from 28 schools, who use Amrita Learning, an ITS developed by Amrita University. There were 211,275 responses obtained in a single academic year. The performance of three knowledge tracing approaches were compared using this dataset: standard Bayesian Knowledge Tracing, Feature-Aware Student Knowledge Tracing (FAST) and Hidden Markov Model (HMM). We find that the HMM approach marginally outperforms the other two methods.