09-20, 14:10–14:45 (Europe/Amsterdam), Van Gogh
Discover the transformative power of SHAP values in machine learning as we decode complex insights into actionable information for stakeholder involvement. Through a unique blend of feature attribution, dimensionality reduction, and clustering, we uncover the crucial drivers behind model predictions, enabling active participation in model co-design. Join us for an engaging session to explore practical case studies and share invaluable lessons for leveraging SHAP values effectively.
Though SHAP values have been part of the machine learning landscape for quite some time, it seems we've only been scratching the surface. Many discussions tend to stay in the realm of numbers and basic visualizations, leaving much of SHAP's potential unexplored.
In this illuminating talk, I will take you on a journey through the co-design of ML models with non-experts by decoding the hidden structure behind SHAP values into something they can grasp. This involves cutting-edge techniques for feature attribution, dimensionality reduction, and clustering to uncover the key decisions driving model predictions. Our approach involves using SHAP values to carve out meaningful subgroups within our data. By condensing these multidimensional values into a more digestible two-dimensional space using UMAP, we're able to unveil hidden patterns with greater clarity. And with the help of clustering methods, we're revealing subgroups within the SHAP value space, each with its own set of clear decision rules. But if that's not enough to engage non-ML experts, we're elevating our approach by visually mapping out these subgroups geographically, adding an extra layer of understanding. Our approach is brought to life through two captivating case studies in the realm of earth sciences: one delving into the assessment of ecological quality of streams, and the other uncovering regions' susceptibility to wildfires.
In essence, our mission is to make SHAP values more accessible and actionable, empowering non-ML experts to gain sufficient insight into the inner workings of the model so that they feel confident to get actively involved in model building and evaluation. Indeed, a challenge for us lies in translating their insights into actionable steps in feature engineering, model architecture, and other technical aspects of model building.
Hans is a senior data scientist. He currently works at Deltares, the Dutch research institute for water and subsurface. He is passionate about actively involving stakeholders in ML model development and creating data-driven models for earth sciences with sufficient extrapolation skill.