JMIR Publications

ABSTRACT

Background:

Keystroke dynamics on smartphones have emerged as a promising form of passive digital biomarker. While prior studies have explored their utility in several diseases and disorders, relatively few have examined how these dynamics change systematically with chronological age in the general population.

Objective:

This study aims to investigate age-related patterns in mobile keystroke dynamics, with a particular focus on temporal variations throughout the day. By identifying behavioral signatures associated with different age groups, we further assess whether AI-based models can accurately estimate chronological age using passively collected keystroke data.

Methods:

We collected smartphone typing data from 177 healthy adults, aged 19–58, over an average of 25 weeks using a custom Android keyboard app. From these data, we extracted 43 keystroke features across categories of typing speed, frequency, and temporal variability. Weekly feature vectors were constructed at multiple temporal resolutions, such as weekly, daily, and 6-hourly medians, and eight AI models were trained to estimate age. A custom loss function was designed to penalize large intra-subject prediction deviations, and feature importance was quantitatively analyzed.

Results:

Descriptive analyses revealed clear age-related differences in typing patterns, with younger participants showing faster and more frequent typing. The LSTM model using 6-hourly-median features achieved the best age estimation performance (MAE 3.69 years, R² 0.71). When the customized loss function was applied, the model’s performance further improved to an MAE of 3.60, with intra-subject variability in estimated ages reduced by 7.8%. Notably, feature importance analysis suggested that the early morning and late evening periods may carry more age-discriminative keystroke patterns.

Conclusions:

Our findings demonstrated that smartphone keystroke dynamics reflect age-sensitive behavioral patterns, particularly when analyzed with fine-grained temporal resolution. While the primary goal was not age estimation per se, the ability to model these patterns highlights the potential of keystroke dynamics as a passive, unobtrusive behavioral marker for age-related functional characteristics. These insights may inform future applications in digital health, such as age-sensitive personalization or early detection of age-related decline without requiring any active user input.