Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Ensembled Penalized Federated Learning Framework for Falling People Detection

Sizhe Rao, Runqiu Zhang, Sajal Saha, Liang Chen

TL;DR

The paper tackles privacy-preserving fall detection by introducing EPFL, an Ensembled Penalized Federated Learning framework that combines penalized local updates with a Specialized Weighted Aggregation SWA and a dual-model ensemble for personalized inference. It leverages homomorphic encryption via TenSEAL to safeguard data while training LSTM-based sequence models, and it incorporates continual learning through a feedback loop. Empirical results on the LDPA dataset show that EPFL+SWA achieves high Recall and F1 scores while outperforming centralized and standard FL baselines, demonstrating strong practical potential for healthcare deployments. The approach is notable for its robust handling of non IID client data, improved stability across rounds, and a privacy-first path toward scalable, personalized fall detection in distributed settings.

Abstract

Falls among elderly and disabled individuals remain a leading cause of injury and mortality worldwide, necessitating robust, accurate, and privacy-aware fall detection systems. Traditional fall detection approaches, whether centralized or point-wise, often struggle with key challenges such as limited generalizability, data privacy concerns, and variability in individual movement behaviors. To address these limitations, we propose EPFL-an Ensembled Penalized Federated Learning framework that integrates continual learning, personalized modeling, and a novel Specialized Weighted Aggregation (SWA) strategy. EPFL leverages wearable sensor data to capture sequential motion patterns while preserving user privacy through homomorphic encryption and federated training. Unlike existing federated models, EPFL incorporates both penalized local training and ensemble-based inference to improve inter-client consistency and adaptability to behavioral differences. Extensive experiments on a benchmark fall detection dataset demonstrate the effectiveness of our approach, achieving a Recall of 88.31 percent and an F1-score of 89.94 percent, significantly outperforming both centralized and baseline models. This work presents a scalable, secure, and accurate solution for real-world fall detection in healthcare settings, with strong potential for continuous improvement via its adaptive feedback mechanism.

An Ensembled Penalized Federated Learning Framework for Falling People Detection

TL;DR

The paper tackles privacy-preserving fall detection by introducing EPFL, an Ensembled Penalized Federated Learning framework that combines penalized local updates with a Specialized Weighted Aggregation SWA and a dual-model ensemble for personalized inference. It leverages homomorphic encryption via TenSEAL to safeguard data while training LSTM-based sequence models, and it incorporates continual learning through a feedback loop. Empirical results on the LDPA dataset show that EPFL+SWA achieves high Recall and F1 scores while outperforming centralized and standard FL baselines, demonstrating strong practical potential for healthcare deployments. The approach is notable for its robust handling of non IID client data, improved stability across rounds, and a privacy-first path toward scalable, personalized fall detection in distributed settings.

Abstract

Falls among elderly and disabled individuals remain a leading cause of injury and mortality worldwide, necessitating robust, accurate, and privacy-aware fall detection systems. Traditional fall detection approaches, whether centralized or point-wise, often struggle with key challenges such as limited generalizability, data privacy concerns, and variability in individual movement behaviors. To address these limitations, we propose EPFL-an Ensembled Penalized Federated Learning framework that integrates continual learning, personalized modeling, and a novel Specialized Weighted Aggregation (SWA) strategy. EPFL leverages wearable sensor data to capture sequential motion patterns while preserving user privacy through homomorphic encryption and federated training. Unlike existing federated models, EPFL incorporates both penalized local training and ensemble-based inference to improve inter-client consistency and adaptability to behavioral differences. Extensive experiments on a benchmark fall detection dataset demonstrate the effectiveness of our approach, achieving a Recall of 88.31 percent and an F1-score of 89.94 percent, significantly outperforming both centralized and baseline models. This work presents a scalable, secure, and accurate solution for real-world fall detection in healthcare settings, with strong potential for continuous improvement via its adaptive feedback mechanism.
Paper Structure (34 sections, 17 equations, 3 figures, 5 tables)

This paper contains 34 sections, 17 equations, 3 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Data Preprocessing and Baseline Models
  4. Time-Series Alignment and Segmentation
  5. Handling Imbalanced Data
  6. Data Standardization
  7. Baseline Model Description
  8. Supervised Binary Classification Models
  9. Unsupervised Outlier Detection Models
  10. Model Selection and Tuning
  11. Proposed Ensembled Penalized Federated Learning (EPFL) Framework
  12. Data Encryption
  13. Client and Global Model Architecture
  14. Proposed Specialized Weighted Aggregation (SWA)
  15. FedNova Normalization
  16. ...and 19 more sections

Figures (3)

  • Figure 1: Workflow for Federated Learningb38
  • Figure 2: Illustration of the proposed model architecture. (a) Describes the complete federated learning workflow with ensemble and penalization components. (b) Shows the internal architecture of the client-side sequential detection model.
  • Figure 3: Training and validation loss over epochs for the sequential LSTM model. The plot demonstrates stable convergence, with the test loss flattening around epoch 45, indicating optimal early stopping.