The settleability of granular sludge is a crucial factor in maintaining an adequate amount of granules in the wastewater treatment process (WWTP). However, Accurate measurement of settleability is challenging due to the complex interactions in WWTP. To address this issue, an intelligent strategy for controlling the settleability of granular sludge is required. This study developed a machine learning (ML)-based model to predict the sludge volume index (SVI30). Three ML models, namely artificial neural networks (ANNs), random forest (RF), and support vector machine (SVM) models, were employed with an ANNs-based mixed liquor suspended solids (MLSS) soft sensor, which served as input data for predicting SVI30 using ML. The combination of the soft sensor’s output and ANNs yielded the highest performance with R2 and mean absolute error (MAE) of 0.8946 and 5.5 mL/g, respectively. The Shapely Additive Explanations (SHAP) analysis revealed that MLSS, salinity, glucose loading rate, and pH were significant factors influencing SVI30. The surrogate model-based optimization strategy of SVI30 was conducted using core control parameters (CCPs), namely glucose loading rate, effluent pH, and salinity. It was suggested that a glucose loading rate over 3.0 kg-N/m3-d and an effluent pH exceeding 9.0 can deteriorate the SVI30. To the best of our knowledge, this is the first study on the surrogate model-based suggestion of the management strategy in the granular sludge process under saline denitrification conditions.