Soil pollution poses a significant challenge globally, affecting both environmental and human health. Traditional methods for predicting soil contamination are limited by the complex interplay of various factors and historical data constraints. This study aims to enhance soil pollution prediction by integrating expert-defined risk zones with advanced machine learning techniques, using the Netherlands as a case study. The research evaluates the impact of expert knowledge on predictive performance through a systematic approach involving data preparation, model structuring, evaluation and interpretation. The findings reveal that while expert-defined risk zones provide some value, their overall contribution to model performance is limited compared to the inherent predictive power of temporal and spatial features.