Mining smartphone generated data for user action recognition – Preliminary assessment

Jakub Fijałkowski , Maria Ganzha , Marcin Paprzycki , Stefka Fidanova , Ivan Lirkov , Costin Badica , Miriana Ivanovic

Abstract

Smartphones became everyday companions of humans. Everyone has a smartphone in their pocket or bag and use it on daily basis. Phones are being used for buying tickets, tracking everyday activities (like working out, running, etc.) and playing games. Unfortunately, most of the applications require manual interactions (e.g. user needs to mark beginning and end of activity) in order to work properly, or need to sacrifice correctness for ease of use (e.g. activity-based games assume that users really perform the activity). Programs can greatly benefit from automatic activity and transportation mode (being some kind of activity) recognition. Current approaches for activity/transportation mode recognition rely on both sensor (accelerometer, gyroscope, magnetometer, etc.) and GPS data, which requires substantial amount of energy to collect and computing power to process. That, in turn, makes them unsuitable to run even on current generation of phones. This paper presents a preliminary result of using just raw sensor data and deep learning techniques for a transportation mode detection, in real-time, directly on a phone. The work tries to balance sensor power consumption and computational requirements with correctness and response time. We also compare results of application of recurrent neural networks, being one of the most powerful means to process time series data, with more traditional approaches (decision trees, SVNs, etc.). Finally, we present approaches that leverage domain knowledge in order to make classifiers more reliable and requiring less processing power (and less energy).
Author Jakub Fijałkowski
Jakub Fijałkowski,,
-
, Maria Ganzha (FMIS / DAICM) - Systems Research Institute (IBS PAN) [Polska Akademia Nauk (PAN)]
Maria Ganzha,,
- Department of Artificial Intelligence and Computational Methods
, Marcin Paprzycki - Systems Research Institute. Polish Academy of Science
Marcin Paprzycki ,,
-
, Stefka Fidanova
Stefka Fidanova,,
-
, Ivan Lirkov
Ivan Lirkov,,
-
, Costin Badica
Costin Badica,,
-
, Miriana Ivanovic
Miriana Ivanovic,,
-
Pages090001:1-090001:18
Publication size in sheets0.3
Book Todorov Michail D. (eds.): Application of Mathematics in Technical and Natural Sciences, AIP Conference Proceedings, vol. 2025, 2018, AMER INST PHYSICS, ISBN 978-0-7354-1745-8
Keywords in Polishanaliza danych, uczenie maszynowe, drzewa decyzyjne, sieci LSTM, CNN, SVN
Keywords in Englishdata analysis, machine learning, LSTM, CNN
Abstract in PolishW artykule zostały omówione wyniki wykorzystania danych, zbieranych z sensorów na smartfonach do wykrywania czy użytkownik jedzie rowerem. Takie rozpoznawanie powinno odbywać się w czasie rzeczywistym (lub prawie rzeczywistym). Poważnym ograniczeniem do takiego wykrywania było wymaganie oszczędzanie energii, co ograniczało lub prawie wykluczało możliwości zbierania danych GPS.
DOIDOI:10.1063/1.5064928
URL https://aip.scitation.org/toc/apc/2025/1?expanded=2025
Languageen angielski
Score (nominal)0
Citation count*
Cite
Share Share

Get link to the record


* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.
Back