Part of the streaming anomaly detection benchmark dataset NAB has keyboard typing data (amongst other interesting sources) [1]. Comparing the algorithms would be quite interesting; the algorithm in [2] is unsupervised and doesn't need training data. The benchmark dataset and algorithms are open-source.

https://github.com/numenta/NAB/tree/master/data#real-data

Ahmad & Lavin, Neurocomputing 2016, https://www.sciencedirect.com/science/article/pii/S092523121...