Groundwater contamination by heavy metals presents a major environmental threat with serious implications for public health and resource sustainability. This study proposes a novel deep learning-based data fusion framework to predict heavy metal contamination index in groundwater, focusing specifically on Manganese (Mn), Iron (Fe), Arsenic (As), and Lead (Pb)—elements found to exceed World Health Organization (WHO) permissible limits in the Gultepe-Zarrinabad sub-basin, Zanjan, Iran. Five widely used water contamination indices (e.g., EHCI, HPI, HEI, MI, and CI) were integrated into a unified composite metric using a customized root-based data fusion and normalization approach. This fused index was modeled using a Deep Neural Network (DNN) and benchmarked against traditional machine learning models—Decision Tree (DT), k-Nearest Neighbors (KNN), and Artificial Neural Network (ANN). The DNN model achieved superior predictive accuracy (R2 = 0.98), with minimal error (RMSE and MAE = 0.01) and excellent generalization capacity, outperforming all other models. This study marks a successful application of a fusion-based DNN approach for comprehensive groundwater heavy metal assessment, demonstrating its strong potential to support AI-enabled environmental monitoring and sustainable water resource management.