RNNs on Monitoring Physical Activity Energy Expenditure in Older People (2006.01169v2)

Abstract: Through the quantification of physical activity energy expenditure (PAEE), health care monitoring has the potential to stimulate vital and healthy ageing, inducing behavioural changes in older people and linking these to personal health gains. To be able to measure PAEE in a monitoring environment, methods from wearable accelerometers have been developed, however, mainly targeted towards younger people. Since elderly subjects differ in energy requirements and range of physical activities, the current models may not be suitable for estimating PAEE among the elderly. Because past activities influence present PAEE, we propose a modeling approach known for its ability to model sequential data, the Recurrent Neural Network (RNN). To train the RNN for an elderly population, we used the GOTOV dataset with 34 healthy participants of 60 years and older (mean 65 years old), performing 16 different activities. We used accelerometers placed on wrist and ankle, and measurements of energy counts by means of indirect calorimetry. After optimization, we propose an architecture consisting of an RNN with 3 GRU layers and a feedforward network combining both accelerometer and participant-level data. In this paper, we describe our efforts to go beyond the standard facilities of a GRU-based RNN, with the aim of achieving accuracy surpassing the state of the art. These efforts include switching aggregation function from mean to dispersion measures (SD, IQR, ...), combining temporal and static data (person-specific details such as age, weight, BMI) and adding symbolic activity data as predicted by a previously trained ML model. The resulting architecture manages to increase its performance by approximatelly 10% while decreasing training input by a factor of 10. It can thus be employed to investigate associations of PAEE with vitality parameters related to metabolic and cognitive health and mental well-being.

• The paper introduces an RNN model that integrates accelerometer and participant data to predict PAEE in older adults with enhanced efficiency.
• The study employs dispersion measures and a personalized data approach via LOSO-CV to significantly reduce training input and improve model robustness.
• The model achieved an R-squared of 0.55 for breath-by-breath and 0.80 for one-minute aggregations, underscoring its practical utility in health monitoring.

An Expert Overview of "RNNs on Monitoring Physical Activity Energy Expenditure in Older People"

The paper "RNNs on Monitoring Physical Activity Energy Expenditure in Older People" discusses the development and validation of a model for estimating physical activity energy expenditure (PAEE) in older adults using recurrent neural networks (RNNs). This paper addresses an identified gap in existing methodologies that primarily focus on younger individuals, failing to account for the differing physical activity characteristics and energy requirements of the elderly.

The authors propose a novel approach leveraging RNNs, known for their capacity to model sequential data, to estimate PAEE from accelerometer data collected from older adults. The research uses the GOTOV dataset, which includes data from 34 participants aged 60 and above, and encompasses 16 different activities. The RNN architecture designed incorporates three Gated Recurrent Unit (GRU) layers and a feedforward network that integrates both accelerometry and participant-level data.

Key Contributions and Methodologies

1. Data Aggregation and Model Architecture:
   • The model transitions from traditional averaging to using dispersion measures (such as standard deviation) for data aggregation. This change enhances model performance, reducing training input by a factor of ten without sacrificing accuracy.
   • The integration of participant-level data (age, weight, BMI) with temporal accelerometer data leads to improved PAEE predictions, indicating the importance of personalized data in such models.
2. Experimental Setup and Evaluation:
   • The research employs Leave One Subject Out Cross-Validation (LOSO-CV) to evaluate the model, ensuring robustness when generalized to new subjects.
   • By combining accelerometer sequences with demographic data, the model's performance improved significantly, highlighting the necessity of individualized modeling in PAEE estimation for older populations.
3. Numerical Results and Performance:
   • The optimal RNN configuration achieved an R-squared value of 0.55 for breath-by-breath PAEE predictions, improving to 0.80 for one-minute aggregations.
   • The model manages to provide competitive estimates compared to traditional methods, while simultaneously simplifying the data processing pipeline by minimizing the need for elaborate feature construction and reducing computational costs.

Implications and Future Directions

The implications of this paper are both practical and theoretical. Practically, the proposed RNN model demonstrates potential for real-world applications in monitoring and promoting physical activity among older populations. It emphasizes the efficacy of combining sensor-based data with individual demographics for tailored health monitoring. Theoretically, the research opens avenues for integrating more complex personalized data streams into machine learning models for health monitoring, suggesting broader applications beyond the scope of PAEE.

Future research could further refine these methods by exploring diverse forms of health-related data or integrating more advanced neural network architectures to enhance prediction accuracy. Additionally, applying this model in longitudinal studies could provide insights into the long-term effects of physical activity interventions on health outcomes in older adults.

In conclusion, this paper makes a substantial contribution to the modeling of PAEE in the elderly using RNNs. It sets a foundation for further advancements in health monitoring technologies geared toward the unique requirements of older populations.