Journal DOI 10.68956/journal.human-performance-futures

Creative Commons source reading · Data quality and algorithm evaluation

Application of Machine Learning Approach to Classify Human Activity Level Based on Lifelog Data

Si-Hwa Jeong; Woomin Nam; Keon Chul Park

Originally published in Sensors (Basel) · 2026-03-04

From the article

Selected source text. JournalOS removes markup and shortens excerpts where indicated. The full article provides the complete context.

Abstract

The present paper provides a human activity-level classification model based on the patient’s lifelog collected from wearable devices. During about two months, the heart rate, step count, and calorie consumption for a total of 182 patients were collected from a wearable device. Using the lifelog data, the machine learning models were developed to classify the physical activity status of patients into five levels. Three types of wearable data with heart rate, step count, and calorie consumption were pre-processed as integrated data in time series. A total of 80% of the integrated data was used as the training dataset, and the remaining 20% was used as the test dataset. Sixteen algorithms were evaluated, including 12 traditional machine learning models (SVM, KNN, RF, etc.) and 4 deep learning models (CNN, RNN, etc.), and…

4. Results

This study suggested machine learning algorithms for classifying human activity levels by applying Python 3.8 and the Scikit-learn library. Section 4.1 summarizes the performance evaluation results and optimal parameters for sixteen algorithms. These models were designed to classify the physical activity status based on patients’ step counts and heart rate data over time. From this comprehensive evaluation, the optimal algorithm was identified and selected for further analysis. Section 4.2 presents the prediction performance evaluation when new patient data are input into the final machine learning models.

4.4. Ablation Study Results: Contribution of Clinical Features

The ablation study results on the test set demonstrated that the integration of clinical features (Model B) led to improved performance in ensemble-based machine learning models. XGBoost showed an increase in accuracy from 0.9440 to 0.9456 and F1-score from 0.9429 to 0.9447. Hard Voting similarly improved its accuracy from 0.9438 to 0.9452. In contrast, the impact on deep learning models was inconsistent; while RNN’s performance slightly declined from 0.9466 to 0.9447, the GRU showed a minor improvement from 0.9433 to 0.9440. These results indicate that while clinical data enhance the predictive power of ensemble-based ML models, its impact on sequential deep learning architectures varies depending on the specific model structure. Furthermore, although the SHAP analysis in Section 4.3 suggested that clinical variables had minimal…

Editorial reading guide

  • What data created and tested the model, and was evaluation independent?
  • What is the cost of a false positive, false negative or out-of-distribution decision?