Identification of human behavior in accelerometer data from interactive toys by applying AutoML.
Eddy Sánchez-DelaCruz1, Cecilia-Irene Loeza-Mejía1, Irahan-Otoniel José-Guzmán1
1Artificial Intelligence Lab., Tecnológico Nacional de México/Instituto Tecnológico Superior de Misantla, Misantla, Veracruz, Mexico.
Physiology & Behavior
|September 24, 2025
Summary
Automated machine learning (AutoML) accurately classifies human behaviors from accelerometer data. This approach achieved 94.4% precision, outperforming traditional methods for analyzing interactions.
Area of Science:
- Human-Computer Interaction
- Machine Learning Applications
- Behavioral Analysis
Background:
- Human behavior is integral to identity, culture, and health, making its study crucial for various research fields.
- Analyzing interactions via sensors provides insights into human behavior.
- This study focuses on classifying behaviors from accelerometer data.
Purpose of the Study:
- To classify human behaviors using accelerometer data from interactive toys.
- To compare the performance of various machine learning algorithms against an AutoML approach.
Main Methods:
- Evaluated machine learning algorithms: Bayes Network, Multinomial Logistic Regression, Multi-layer Perceptron, Naïve Bayes, and RIPPER.
- Applied an Automated Machine Learning (AutoML) approach for automated model and hyperparameter selection.
- Analyzed 8,946 behavior records across five classes: drop, hit, pickup, shake, and throw.
Main Results:
- AutoML significantly outperformed traditional machine learning classifiers.
- Achieved a precision of 94.4% in behavior classification.
- Obtained a Receiver Operating Characteristic (ROC) area of 0.992.
Conclusions:
- AutoML demonstrates high effectiveness in accurately identifying human behaviors.
- Accelerometer data combined with AutoML offers a robust method for behavioral analysis in interactive systems.
