SapiMouse - a new dataset for Mouse Dynamics (2020)
Repository for paper: SapiMouse: Mouse Dynamics-based User Authentication Using Deep Feature Learning
- Data collection software: https://mousedynamicsdatalogger.netlify.app
- Raw data: https://ms.sapientia.ro/~manyi/sapimouse/sapimouse.html
- First session: S3 (3 minutes)
- Second session: S1 (1 minute)
- 120 subjects (92 male, 28 female)
- log file lines: [timestamp, button, state, x, y]
- step 1: first order differences (absolute value) - |dx|,|dy|
- step 2: segmentation into fixed-sized blocks (128)
-
split the dataset into 2 subsets * subset 1: subjects 1 .. 72 (60%) * subset 2: subjects 73 ..120 (40%)
-
step 1: feature learning (training a fully convolutional neural network, see TRAINED_MODELS) using subset 1
-
step 2: creating a One-class SVM model for each subject from subset 2
- Margit Antal, Norbert Fejer, Krisztian Buza (2021), SapiMouse: Mouse Dynamics-based User Authentication Using Deep Feature Learning, May 19-21, 2021, LINK.
- M. Antal, K. Buza and N. Fejer (2021), "SapiAgent: A Bot Based on Deep Learning to Generate Human-Like Mouse Trajectories," in IEEE Access, vol. 9, pp. 124396-124408, 2021, doi: 10.1109/ACCESS.2021.3111098, IF: 3.367, LINK.