Evaluating techniques based on supervised learning methods in casas kyoto dataset for human activity recognition

One of the technical aspects that contribute to improving the quality of life for older adults is the automation of physical spaces using sensors and actuators, which facilitates the performance of their daily activities. The interaction between individuals and their environment enables the detectio...

Full description

Autores:
García Restrepo, Johanna Karinna
Ariza Colpas, Paola Patricia
Butt Aziz, Shariq
Piñeres Melo, Marlon Alberto
Naz, Sumera
De la hoz Franco, Emiro
Tipo de recurso:
Article of investigation
Fecha de publicación:
2023
Institución:
Corporación Universidad de la Costa
Repositorio:
REDICUC - Repositorio CUC
Idioma:
eng
OAI Identifier:
oai:repositorio.cuc.edu.co:11323/13664
Acceso en línea:
https://hdl.handle.net/11323/13664
https://repositorio.cuc.edu.co/
Palabra clave:
Activity of Daily Living
Classification Methods
Human Activity Recognition
Selection Methods
Smart home
Rights
closedAccess
License
Atribución 4.0 Internacional (CC BY 4.0)
Description
Summary:One of the technical aspects that contribute to improving the quality of life for older adults is the automation of physical spaces using sensors and actuators, which facilitates the performance of their daily activities. The interaction between individuals and their environment enables the detection of abnormal patterns that may arise from a decline in their cognitive abilities. In this study, we evaluate the CASAS Kyoto dataset from WSU University, which provides information on the daily living activities of individuals within an indoor environment. We developed a model to predict activities such as Cleaning, Cooking, Eating, Washing hands, and Phone Call. A novel approach is proposed, which involves preprocessing and segmenting the dataset using sliding windows. Furthermore, we conducted experiments with various classifiers to determine the optimal choice for the model. The final model utilizes the regression classification technique and is trained on a reduced dataset containing only 5 features. It achieves outstanding results, with a Recall of 99.80% and a ROC area of 100%.