<p>An innovative methodology for predicting gradation stability using integrated machine learning technologies is introduced. Current geometric criteria for suffusion assessment rely on a limited set of characteristic particle sizes, which results in a loss of detailed gradation information embedded in grading curves. This study proposes a new framework for evaluating the suffusion sensitivity through predicting the gradation stability of granular soil with a specified grading curve. Two distinct integrated machine learning models are developed to quantitatively assess soil internal stability. The predicted results and performance analysis demonstrate that the PCA-SVM model achieves superior classification accuracy for internal stability, while the PCA-ANN exhibits enhanced predictive capability in estimating the probability of internal stability within the given dataset. The proposed methodology provides a novel application for investigating the relationship between gradation characteristics and stability. This study will facilitate further research on establishing the accurate gradation stability criteria and predicting the soil suffusion sensitivity.</p> Graphical Abstract <p></p>