Prepoznavanje obrokov in ugrizov s strojnim učenjem na podlagi meritev pametne ure : magistrsko delo

Jordan, Marko (author), Luštrek, Mitja (mentor), Košir, Tomaž (comentor)

Magistrska naloga predstavi pristop zaznavanja obrokov in ugrizov. Podatki so bili pridobljeni v eksperimentu, v katerem je sodelovalo 12 posameznikov, ki so s pomočjo žiroskopa in pospeškomera, nameščenih v pameti uri, merili premike zapestja v svojem običajnem življenju. Skupno so posneli za 570 ur podatkov in pri tem zajeli 169 obrokov. Glavni del algoritma za zaznavanje obrokov in ugrizov temelji na uporabi globokega učenja. Pri rezultatih, dobljenih s pomočjo prečnega preverjanja, je mera F1 v primeru zaznavanja obrokov znašala 0,91, medtem ko je pri zaznavanju ugrizov znašala 0,83. To kaže, da se je algoritem relativno uspešno naučil prepoznavati obroke v običajnem življenju posameznikov, medtem ko se je prepoznavanje ugrizov izkazalo za večji izziv.

• Language: Slovenian
• Keywords: strojno učenje, globoko učenje, prepoznavanje prehranjevanja
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FMF - Faculty of Mathematics and Physics
• Year: 2024
• PID: 20.500.12556/RUL-154905
• COBISS.SI-ID: 187908355
• Publication date in RUL: 08.03.2024

JORDAN, Marko, 2024, Prepoznavanje obrokov in ugrizov s strojnim učenjem na podlagi meritev pametne ure : magistrsko delo [online]. Master’s thesis. [Accessed 5 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=154905

Secondary language

• Language: English
• Title: Predicting meals and bites with machine learning based on measurements of a smartwatch

Abstract

The master’s thesis proposes an approach of predicting meals and bites. The data was obtained in the wild, in an experiment in which 12 participants measured the movements of their wrists via a gyroscope and an accelerometer installed in a smartwatch. The data includes 169 meals and has the total duration of 570 hours. The main part of the algorithm for the predictions of meals and bites is based on deep learning. The results obtained with a cross-validation show that the F1 score of the predictions of meals equals 0.91, while the F1 score of the predictions of bites equals 0.83. This shows that the algorithm relatively successfully learned to recognize meals in the wild, while the predictions of bites proved to be more challenging.

• Keywords: machine learning, deep learning, eating recognition