Context

A consortium led by Groupe UP and Institut Mines-Télécom is looking to design an alert platform for homecare structures, capable of automatically detecting beneficiaries at increasing risk of loss of autonomy. The data collected, from multiple aid organizations, is of uneven quality, widely scattered and includes sensitive personal information subject to strict anonymization requirements.

The AI part of the project is being carried out by Datalchemy.

Completed work

  • Mapping, cleaning and enrichment of raw data sets.
  • Design and supply of a centralized data architecture.
  • Data anonymization and audit of direct or indirect identification risks.
  • Construction of datasets and development of neural network models.
  • Model training and performance validation on real cases.

Results

  • Despite the scarcity and poor overall quality of the data, the project was a success, providing homecare structures with better information on the health status of their beneficiaries.
  • The consortium plans to industrialize the project.