Skip to content

gsi-upm/fl_for_frailty

Repository files navigation

A Federated Learning Approach for Frailty Detection

This repository contains the experiments related to a proof of concept of the application of Federated Learning techniques to the problem of frailty screening or early detection.

To properly run this repository you should have access to the data of wave 6 of ELSA, the English Longitudinal Study of Ageing (see portal https://www.elsa-project.ac.uk/). Spcecifically, you will need the files "wave_6_elsa_data_v2.tab" and "wave_6_elsa_nurse_data_v2.tab". Once you have the data, refer to the following repository to gerenerate a frailty label following Fried's Frailty Phenotype definition: https://github.com/gsi-upm/FRELSA .

Alternatively, the file "frailty_levels.csv" contains the correspondence between the ELSA participants' unique index ("idauniq") and their frailty level (0 = "non-frail"; 1 = "pre-frail"; 2 = "frail"). You need to merge the two datasets contained in the files by the participants' unique index ("idauniq"), and then label them with the "frailty_levels.csv" labels.

The dataset is divided into nine regions, with the distribution shown in the following image:

In the "Regional_and_Global_Training" notebook, specific Logistic Regression and MLP architectures are trained, both regionally and on the whole dataset. The results are then compared to a federated strategy in which the nine regions cooperate to improve the performance, imagining they are not allowed to share data.

Execute the command

flwr run .

to launch the federated experiments on the same regional distribution.

Results show that the federated strategy is equivalent to the global training.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published