I'm Klest Dedja,

Passionate about mathematics, my background bridges Applied Mathematics and Machine Learning.

During my PhD at KU Leuven under the supervision of Prof. Celine Vens I specialized Explainable AI for Survival Analysis tasks (time-to-event in the presence of partial information) with applications in healthcare (Multiple Sclerosis, kidney function, etc.). I have explored Random Forests extensively, as their inherent structure provides both flexibility in predictive modeling and potential pathways to interpretability.

Additionally, I reseached into the intersection of Active Learning and Survival Analysis, with findings and tested approaches shown in this other project. You can find my PhD dissertation here.

• I have worked at Predikt, a young, innovative start-up dedicated to advancing time-series forecasting for CFOs and finance leaders. Predikt is developing AI-driven tools that bring confidence and trust to forecasting processes, helping CFOs and finance leaders make more informed decisions.

This combination of deep theoretical expertise, experience in the startup world, and 7+ years of programming makes me a strong and curious Data Scientist.

• I am maintaining a project from my PhD years, namely BELLATREX: an open-access package designed to support adoption and transparency of Random Forest models for several prediction tasks: binary classification, regression, survival-analysis, multi-lablel classification, and multi-target regression.

  Do you like BELLATREX? I am looking for collaborators to make BELLATREX better! If you have fresh ideas, feature requests, or are interested in contributing to new functionalities, I’d love to connect 😊. Keep an eye on the repository and don't forget to add a ⭐️

• Another project involves extending SHAP explanatory toolbox to time-to-event data, with a focus on explaining feature importance across several time intervals through IntervalSHAP. This method unlocks insights that might otherwise go unnoticed, and is a fast and lean alternative to SurvSHAP(t).

• To be released to the public upon acceptance of the related paper: EDGEHOG, a tool for automatic directionality dispersion estimate, using a classical Computer vision approach

You can find me on LinkedIn: Klest Dedja.