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

Lifelog data from wearable devices collected over approximately two months was used to develop machine learning models for classifying physical activity into five levels.

• Data collected from 182 patients over approximately two months
• Three types of wearable data were used: heart rate, step count, and calorie consumption
• Physical activity status was classified into five levels
• Data were pre-processed as integrated data in time series

Sixteen machine learning algorithms were evaluated, including both traditional and deep learning models.

• 12 traditional machine learning models were evaluated, including SVM, KNN, and RF
• 4 deep learning models were evaluated, including CNN and RNN
• Cross-validation was performed by dividing the training dataset into 5 folds
• Parameters were tuned to derive models with optimal parameters

An 80/20 train-test split was applied to the integrated dataset for model development and evaluation.

• 80% of the integrated data was used as the training dataset
• The remaining 20% was used as the test dataset
• Final model performance was evaluated with new patient lifelog data

Classification of human activity level based on heart rate and step count can be performed with high accuracy.

• Final models were evaluated using new patient lifelog data
• Heart rate and step count were identified as the key features for high-accuracy classification
• Calorie consumption was also included as one of the three input data types
• The paper states classification 'can be performed with high accuracy' based on heart rate and step count