An accurate assessment of leachate levels necessitates the integration of various parameters. Traditional geophysical prospecting methods often lack measurable accuracy because they focus on individual parameters rather than effectively integrating data. This may lead to inconsistent estimates of leachate depth and make the evaluation of prediction reliability challenging. In this study, we exploit hard and soft cluster analyses to improve the effectiveness of geoelectrical methods in identifying the extent of leachate accumulation zones. A machine learning-based approach employing hard clustering on resistivity and induced polarization data was recently proposed to obtain integrated model sections that highlight leachate accumulation zones in municipal solid waste landfills. In those models, areas with different colours represent areas characterized by specific ranges of values of the considered physical quantities and have strictly defined boundaries. This is an intrinsic limitation of hard clustering that carries out cluster assignments without providing an assessment of the reliability of the reconstruction. In contrast, soft clustering approaches provide estimation of the cluster membership that allows a refinement of cluster boundaries, improving the identification of groups in the data. We apply hard and soft cluster analyses to geoelectrical data for detecting leachate accumulation zones in a landfill located on a steep slope in Central Italy. There, leachate may not only contaminate groundwater but also trigger instability phenomena. Among the different clustering algorithms, we selected K-means due to its simplicity of implementation, its ability to identify clusters that are both compact and distinct, and its faster performance compared to Fuzzy C-means. The clusters associated with the leachate accumulation zones represent approximately the 11% of total investigated subsoil and are characterized by values of resistivity, chargeability and normalized chargeability in the ranges of about 1.5-5 Ωm, 10-70 mV/V and 4.5-37 mS/m, respectively. Then, we applied the Fuzzy C-means algorithm to obtain the degree of membership of points belonging to such areas and better outline their boundaries. By considering a fuzzy membership greater than 0.5, we achieve an accuracy exceeding 90% in identifying leachate in wells. Furthermore, identifying zones with lower membership we delineate the boundaries of less saturated regions as well as those that are more saturated, providing a reconstruction of potential preferential leachate flows within the waste mass. These findings have important practical implications as they contribute to cost reductions for future drilling and monitoring processes.