PyXAI In-the-Loop

Logic-Based Explanations of Imbalance Price Forecasts Using Boosted Trees

This common work with Jérémie Bottieau and J.-F. Toubeau is published in PSCC 2024.

Due to the necessity to continuously keep a balance between generation and consumption, the real-time trade and pricing of electricity is central in liberalized power systems. In this way, market participants, such as producers and suppliers, trade quantities of energy in forward markets based on their expectation of (future) real-time electricity prices. For each delivery period, if they produce or consume quantities that differ from what they have contracted, imbalance volumes are settled at a real-time price, called the imbalance price in Europe. Depending on the system conditions, imbalance payments may be extremely high, which can lead to bankruptcy for energy traders if they are not appropriately hedged. On the other hand, such high prices may also offer valuable temporal arbitraging opportunities for fast-ramping units.

Using PyXAI, we investigate the driving factors that conduct the prediction model to infer a high or low imbalance price regime.

Improving care interactions (and training) in nursing homes with artificial intelligence

This work is done by Marie Lefelle and Mouny Samy Modeliar and is published in GeroScience.

With an ageing population, nursing homes are set to play an increasingly important role as care facilities for dependent people. To improve the well-being of the elderly, one important perspective is to work on language skills mobilized during care interactions. However, the taboo of dependency, the various scandals involving private nursing home management companies, the expected efficiency of caregivers, and the plurality of care contexts don’t make it easy to monitor such skills. A loophole is by collecting data in situ, while being supervised by language researchers and caregivers specialized in elderly care.

This is the approach we have followed: the data collected was then exploited by means of machine learning models that are capable of providing caregivers with information that is important to the success of care acts. In particular, the results of our research show the importance of some key factors of language-based interactions (while paying attention to diversified situations).

PyXAI applied to our context allows us to better understand the reasons for the success or failure of a care session. This method is highly relevant in the context of caregiving, as it allows us to determine the variables that are key factors: they are discriminating variables that predict the success or failure of care interventions.

Predicting Critical Deterioration of Patients in Emergency Units Using Administrative Health Data

This work in done by Clément Lens, Bilal Majed, Pierre Marquis, Karim Tabia and Romain Wallon. It is published in Artificial Intelligence in Medecine.

Due to increasing pressure on emergency departments, particularly driven by population aging, decision support tools are needed to improve patient prioritization. They propose an AI-based approach to identify patients at risk of critical deterioration at emergency admission using only routinely collected medical-administrative data. The method relies exclusively on PMSI data, mainly diagnoses and procedures, which are widely available in hospital information systems. A boosted tree model is trained to predict deterioration while providing formal explanations. Despite the absence of physiological data, the model achieves strong predictive performance and offers interpretability. A subset of predictions reviewed by physicians did not reveal obvious errors.