With cable equipment being widely deployed, cable faults have threatened the safe operation of the power grid. Traditional operation and maintenance work has difficulty in accurately predicting the current health status of cable insulation. To address this issue, a cable health index prediction method based on an improved genetic algorithm-back propagation (IGA-BP) neural network model is proposed. Since the rate of parameter change in underground cables varies at different aging stages, the method incorporates recent aging trend characteristics into both the fitness function and mutation operator during parameter optimization. By distinguishing individuals based on these aging characteristics, the model enhances both the efficiency of searching for a global optimum and the accuracy of predictions. Experimental results demonstrate that compared to traditional back propagation (BP) and genetic algorithm?back propagation (GA-BP) neural networks, the IGA-BP neural network improves prediction accuracy by 3.68%, achieving 99.39% accuracy in five-fold cross-validation and 95.8% accuracy in a dataset of 15 kV high-voltage cross-linked polyethylene (XLPE) underground cables. The developed model is well-suited for health index prediction as it fully accounts for the historical aging information of cables.