Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Benchmarking Classical, Deep, and Generative Models for Human Activity Recognition

Md Meem Hossain, The Anh Han, Safina Showkat Ara, Zia Ush Shamszaman

TL;DR

This paper addresses HAR by benchmarking classical ML, deep learning, and Restricted Boltzmann Machine methods across five standard datasets (UCI-HAR, OPPORTUNITY, PAMAP2, WISDM, Berkeley MHAD). It evaluates a broad model set including DT, RF, SVM variants, CNN, RNN/LSTM/BiLSTM/GRU, ANN, DBNs, and DBMs using accuracy, precision, recall, and F1-score. Key findings show CNNs consistently outperform others, especially on complex datasets like Berkeley MHAD, while RF excels on smaller datasets and RBMs offer strong feature-learning potential, sometimes approaching deep models. These results provide practical guidance for selecting HAR models based on dataset size, modality, and computational constraints, and highlight RBMs as a viable path for feature learning and cross-domain HAR exploration.

Abstract

Human Activity Recognition (HAR) has gained significant importance with the growing use of sensor-equipped devices and large datasets. This paper evaluates the performance of three categories of models : classical machine learning, deep learning architectures, and Restricted Boltzmann Machines (RBMs) using five key benchmark datasets of HAR (UCI-HAR, OPPORTUNITY, PAMAP2, WISDM, and Berkeley MHAD). We assess various models, including Decision Trees, Random Forests, Convolutional Neural Networks (CNN), and Deep Belief Networks (DBNs), using metrics such as accuracy, precision, recall, and F1-score for a comprehensive comparison. The results show that CNN models offer superior performance across all datasets, especially on the Berkeley MHAD. Classical models like Random Forest do well on smaller datasets but face challenges with larger, more complex data. RBM-based models also show notable potential, particularly for feature learning. This paper offers a detailed comparison to help researchers choose the most suitable model for HAR tasks.

Benchmarking Classical, Deep, and Generative Models for Human Activity Recognition

TL;DR

This paper addresses HAR by benchmarking classical ML, deep learning, and Restricted Boltzmann Machine methods across five standard datasets (UCI-HAR, OPPORTUNITY, PAMAP2, WISDM, Berkeley MHAD). It evaluates a broad model set including DT, RF, SVM variants, CNN, RNN/LSTM/BiLSTM/GRU, ANN, DBNs, and DBMs using accuracy, precision, recall, and F1-score. Key findings show CNNs consistently outperform others, especially on complex datasets like Berkeley MHAD, while RF excels on smaller datasets and RBMs offer strong feature-learning potential, sometimes approaching deep models. These results provide practical guidance for selecting HAR models based on dataset size, modality, and computational constraints, and highlight RBMs as a viable path for feature learning and cross-domain HAR exploration.

Abstract

Human Activity Recognition (HAR) has gained significant importance with the growing use of sensor-equipped devices and large datasets. This paper evaluates the performance of three categories of models : classical machine learning, deep learning architectures, and Restricted Boltzmann Machines (RBMs) using five key benchmark datasets of HAR (UCI-HAR, OPPORTUNITY, PAMAP2, WISDM, and Berkeley MHAD). We assess various models, including Decision Trees, Random Forests, Convolutional Neural Networks (CNN), and Deep Belief Networks (DBNs), using metrics such as accuracy, precision, recall, and F1-score for a comprehensive comparison. The results show that CNN models offer superior performance across all datasets, especially on the Berkeley MHAD. Classical models like Random Forest do well on smaller datasets but face challenges with larger, more complex data. RBM-based models also show notable potential, particularly for feature learning. This paper offers a detailed comparison to help researchers choose the most suitable model for HAR tasks.
Paper Structure (39 sections, 10 equations, 21 figures, 20 tables)

This paper contains 39 sections, 10 equations, 21 figures, 20 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Research Design and Data Collection
  4. Datasets
  5. UCI-HAR
  6. Opportunity
  7. PAMAP 2
  8. WISDM (Wireless Sensor Data Mining)
  9. The Berkeley MHAD (Multimodal Human Action Database)
  10. Models
  11. Decision Tree
  12. Random Forest
  13. Sampling
  14. Logistic regression
  15. Linear SVC
  16. ...and 24 more sections

Figures (21)

  • Figure 1: A general illustration of a decision tree.
  • Figure 2: A visual illustration of the Random Forest
  • Figure 3: A visual illustration of KNN algorithm
  • Figure 4: Visual representation of different layers in a CNN.
  • Figure 5: RNNs utilize their internal state, or memory, to analyze sequences of inputs.
  • ...and 16 more figures