Date of Award

2018

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Abstract

Pinky Sindhu

Loyola University Chicago

TODDLER ACTIVITY RECOGNITION USING MACHINE LEARNING

Toddlers behave differently than adults, to say the least. It is valuable to accurately measure the specific types of physical activity (PA) in toddlers; such information can be analyzed to predict future health prospects in relation to conditions like obesity.

We attached ActiGraph accelerometers to the wrist and waist of toddlers and recorded PAs. Toddlers were videotaped, and their movements were annotated as 20 specific activities. These activities were classified into 3 summary activity intensities including sedentary, light intensity PA (LPA), and moderate to vigorous intensity PA (MVPA).

Automated activity recognition proceeded through a series of machine learning signal processing stages. To train the activity classifier 81 signal processing features were extracted from every two second clips of sensor data. When training on 20 activities, the overall accuracy was 63.8%. When the activities were grouped into 3 intensity levels, the highest accuracy was 73.6%, also using the Random Forest classifier.

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 License.

Share

COinS