Soil stabilization is a widely used technique in geotechnical engineering to improve the strength and durability of weak or problematic soils for construction. This process typically involves adding stabilizing agents like cement, lime, fly ash, or bitumen to enhance soil properties. Traditional methods for evaluating stabilization, such as extensive laboratory testing, are time-consuming and resource-intensive. This study explores the use of Machine Learning (ML) algorithms to predict the effectiveness of various additives in soil stabilization, aiming to streamline the design process and reduce testing. A diverse dataset is compiled from published studies and soil databases, including soil properties, additive types, dosage, curing time, and environmental conditions. ML models, including Support Vector Machines (SVM), Random Forests (RF), Artificial Neural Networks (ANN), and Gradient Boosting (GB), are trained to predict stabilization outcomes like Unconfined Compressive Strength (UCS), cohesion, and soil plasticity. Here RF model is used specifically for predicting UCS values. Model performance is evaluated using metrics such as R2, Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Results show that the RF model achieves 75–80% accuracy on test data, outperforming traditional methods. Feature importance analysis identifies additive type, dosage, curing age, and initial soil composition as key factors influencing stabilization efficiency. This study demonstrates that ML can offer a more efficient, cost-effective alternative to conventional techniques, enabling better decision-making in soil stabilization for construction projects.