France Travail (ex Pôle Emploi)

Challenge

Pôle Emploi teams collect a lot of data: structured, unstructured, time series, etc. Pôle Emploi needs to use some of this data to provide personalized services while complying with European rules.

The aim? Use anonymized data to feed the Machine Learning model. Avoid seeking new consent by ensuring regulatory compliance. Avoid sensitive data leaks.

‍
Solution

• The avatar personal data anonymization solution made it possible to free up the use of Pôle Emploi data.
• To allow the exploitation of this data, Octopize has deployed its avatar solution on the Pôle Emploi servers (on premise).
• After training, Pôle Emploi employees were then able to anonymize their data and use avatar data to train their machine learning algorithm, in autonomy and in compliance.
• Avatar data stores statistical properties of the original dataset as well as its variability, without compromising the confidentiality individuals at their origin.

Setting up

• Service provision
• Solution deployed on Pôle Emploi servers
• Formation internal teams

Maintaining statistical quality & utility

F1 original score = 0.52

F1 avatar score = 0.52

Original AUC = 0.70

Avatar AUC = 0.69

ROI

• Maintaining the prediction performance of the AI model with very good results, no loss of performance or even increases in some categories
• Data publication in Open Source
• Sharing the informative value of data without compromise
• Role of reinsurance of users In the event of an audit

‍

Additional resource

Find the testimony by Laurent Guinard, head of the Data Service Agency & AI Factory department at France Travail, ex Pôle Emploi (2 min).

As a user of our method of anonymizing personal data, discover:

• His initial challenge #MachineLearning #IA
• The answer to his needs by the method avatar
• His return to the profits drawn ‍