Show/Hide Menu
Hide/Show Apps
Logout
Türkçe
Türkçe
Search
Search
Login
Login
OpenMETU
OpenMETU
About
About
Open Science Policy
Open Science Policy
Open Access Guideline
Open Access Guideline
Postgraduate Thesis Guideline
Postgraduate Thesis Guideline
Communities & Collections
Communities & Collections
Help
Help
Frequently Asked Questions
Frequently Asked Questions
Guides
Guides
Thesis submission
Thesis submission
MS without thesis term project submission
MS without thesis term project submission
Publication submission with DOI
Publication submission with DOI
Publication submission
Publication submission
Supporting Information
Supporting Information
General Information
General Information
Copyright, Embargo and License
Copyright, Embargo and License
Contact us
Contact us
PREDICTING USER TEXT TYPING PERFORMANCE FROM SENSOR DATA AND KEYSTROKES ON MOBILE DEVICES
Download
baris_gokmen_msc_thesis.pdf
Date
2026-9-11
Author
Gökmen, K. Barış
Metadata
Show full item record
Item Usage Stats
71
views
0
downloads
Cite This
Smartphone typing takes place in many everyday situations, but typing in different contexts can lead to situationally induced impairments and disabilities (SIID). Understanding SIID is important for designing better text-entry systems; previous studies have examined the relationship between typing performance and context. However, these studies either use controlled settings or try to identify the user’s activity or context instead of predicting typing performance directly. This thesis examines whether smartphone sensor data can be used to predict continuous typing performance without first classifying the user’s activity or situation. Data were collected in everyday use from 11 participants using their own Android phones and the AWARE Framework. After privacy filtering and event-based segmentation, the dataset contained 21,279 valid typing segments, together with overlapping two- and five-second windows. The main analysis used 93 features from the accelerometer, gyroscope, and gravity sensors. Spearman correlations were used to study the correlation between sensor features and typing performance. Five regression models were tested using repeated personalized 60/20/20 train, validation, and test splits. Leave-One-Participant-Out validation was also used to test performance on unseen users as a cross-user generalized model. The prediction targets were words per minute (WPM), keystrokes per second (KSPS), and keystrokes per character (KSPC). Accelerometer jerk and gyroscope rate-of-change features were often related to higher WPM and KSPS, while gyroscope lag-one autocorrelation was usually linked to lower values. The links with KSPC were weaker and less consistent. In the main five-second personalized evaluation, Extra Trees achieved an average MAE of 10.807 and an R2 of 0.309 for WPM. XGBoost achieved an average MAE of 0.619 and an R2 of 0.469 for KSPS. Crossuser results were weaker, and KSPC could not be predicted reliably. These results show that motion sensors contain useful user-specific information about short-term typing speed, and that estimation of typing performance from the same user, modeled with its data, is indeed possible. However, the models did not generalize strongly to unseen users. The thesis provides a direct sensor-to-typing-performance pipeline and practical evidence for future sensor-aware text-entry systems.
Subject Keywords
Personalized regression
,
Typing-performance prediction
,
mobile-typing
,
In-the-wild sensing
,
Mobile text entry
,
Smartphone sensors
URI
https://hdl.handle.net/11511/121013
Collections
Northern Cyprus Campus, Thesis
Citation Formats
IEEE
ACM
APA
CHICAGO
MLA
BibTeX
K. B. Gökmen, “PREDICTING USER TEXT TYPING PERFORMANCE FROM SENSOR DATA AND KEYSTROKES ON MOBILE DEVICES,” M.S. - Master of Science, Middle East Technical University, 2026.