In-depth Analysis of the Performance and Applications of Classical and Deep Learning Models under the UCI HAR Dataset

Yuqin Yang; Xuxin Wang

doi:10.54254/2755-2721/2026.TJ29093

Applied and Computational EngineeringOpen access

In-depth Analysis of the Performance and Applications of Classical and Deep Learning Models under the UCI HAR Dataset

Research Article

Open Access

In-depth Analysis of the Performance and Applications of Classical and Deep Learning Models under the UCI HAR Dataset

Yuqin Yang ¹, Xuxin Wang ^2*

¹ Eastside Preparatory School, Bellevue, United States

² Hong Kong Polytechnic University, Hong Kong, China

^*Corresponding author: 24110796d@connect.polyu.hk

Published on 5 November 2025

ACE Vol.203

ISSN (Print): 2755-273X

ISSN (Online): 2755-2721

ISBN (Print): 978-1-80590-515-8

ISBN (Online): 978-1-80590-516-5

Download Cover

Abstract

Human Activity Recognition (HAR) is vital in pattern recognition and AI, with applications in smart health and security, but faces challenges like diverse activities and sensor noise, making model selection critical. In this work, we compare classical and deep learning models for HAR using the UCI HAR dataset—including inertial sensor data from 30 volunteers performing six activities. Classical models (linear/RBF SVM, Random Forest, KNN, AdaBoost, Stacking) and deep learning models (CNN, RNN-LSTM) are evaluated on accuracy, macro-F1, and resource metrics. A leakage-free workflow is adopted: GridSearchCV (3-fold cross-validation) tunes hyperparameters, models are retrained on the full training set, and tested independently. Results show linear SVM achieves the best single-model accuracy (96.13%), while Stacking (combining linear SVM, KNN, RF) performs best overall (96.61%). CNN (92.60% accuracy) slightly outperforms RNN-LSTM (90.91%), and KNN uses the least memory. This work provides key insights for HAR model selection (linear SVM as baseline, Stacking for accuracy) and guides future work to reduce false positives, advancing HAR technology.

Keywords:

human activity recognition (HAR), UCI HAR dataset, classical and deep learning models, model performance comparison

View PDF

References

[1]. Hammerla et al. (2016) Deep, Convolutional, and Recurrent Models for Human Activity Recognition using Wearables.

[2]. MDPI Paper (2020) Comparing Human Activity Recognition Models Based on Sensor Data.

[3]. Gaikwad et al. (2024) Benchmarking Classical, Deep, and Generative Models for Human Activity Recognition.

[4]. Ahmad et al. (2020) Human Activity Recognition using Multi‑Head CNN followed by LSTM.

[5]. Mathew et al. (2023) Human activity recognition using Single Frame CNN and ConvLSTM.

[6]. Khan & Hossni (2025) Comparative Analysis of LSTM Models Aided with Attention and Squeeze‑and‑Excitation Blocks for Activity Recognition (Scientific Reports)

References

[1]. Hammerla et al. (2016) Deep, Convolutional, and Recurrent Models for Human Activity Recognition using Wearables.

[2]. MDPI Paper (2020) Comparing Human Activity Recognition Models Based on Sensor Data.

[3]. Gaikwad et al. (2024) Benchmarking Classical, Deep, and Generative Models for Human Activity Recognition.

[4]. Ahmad et al. (2020) Human Activity Recognition using Multi‑Head CNN followed by LSTM.

[5]. Mathew et al. (2023) Human activity recognition using Single Frame CNN and ConvLSTM.

[6]. Khan & Hossni (2025) Comparative Analysis of LSTM Models Aided with Attention and Squeeze‑and‑Excitation Blocks for Activity Recognition (Scientific Reports)

Cite this article

Yang,Y.;Wang,X. (2025). In-depth Analysis of the Performance and Applications of Classical and Deep Learning Models under the UCI HAR Dataset. Applied and Computational Engineering,203,53-64.

Data availability

The datasets used and/or analyzed during the current study will be available from the authors upon reasonable request.

About volume

Volume title: Proceedings of CONF-SPML 2026 Symposium: The 2nd Neural Computing and Applications Workshop 2025

ISBN: 978-1-80590-515-8(Print) / 978-1-80590-516-5(Online)

Editor: Marwan Omar, Guozheng Rao

Conference website: https://www.confspml.org/tianjin.html

Conference date: 21 December 2025

Series: Applied and Computational Engineering

Volume number: Vol.203

ISSN: 2755-2721(Print) / 2755-273X(Online)

© 2024 by the author(s). Licensee EWA Publishing, Oxford, UK. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license. Authors who publish this series agree to the following terms:

1. Authors retain copyright and grant the series right of first publication with the work simultaneously licensed under a Creative Commons Attribution License that allows others to share the work with an acknowledgment of the work's authorship and initial publication in this series.

2. Authors are able to enter into separate, additional contractual arrangements for the non-exclusive distribution of the series's published version of the work (e.g., post it to an institutional repository or publish it in a book), with an acknowledgment of its initial publication in this series.

3. Authors are permitted and encouraged to post their work online (e.g., in institutional repositories or on their website) prior to and during the submission process, as it can lead to productive exchanges, as well as earlier and greater citation of published work (See Open access policy for details).