KDBC-Kollect : Plateforme de Collecte de Données de Dynamique de Frappe au Clavier et Authentification par Apprentissage

National audience This article presents KDBC-Kollect, a platform dedicated to the collection of keystroke dynamics data tailored to the Chadian context. As a result of such Western dependency of installed biometric systems, we gathered 6,000 samples of 100 Chadian users. The benchmarking between 100 fine-tuned TypeNet models reveals an average Equal Error Rate (EER) of 7.83% and 57% of the models can realize an EER ≤5%. The comparison with the GREYC-KeyStroke dataset is significant (p < 0.001) which justifies that contextual adaptation is essential.