Unconfined Compressive Strength (UCS) is usually the significant design criterion for in situ ground improvement works. In practice, Ordinary Portland Cement (OPC) is the most commonly used binder for the stabilisation of soft clays due to its efficiency and ease of availability. However, using OPC emits a large amount of CO2 and also increases the total cost of the project. Therefore, researchers are seeking alternative binders such as granulated blast furnace slag (GGBFS) in conjunction with OPC, which is named Portland blast furnace slag cement (PBFC) or type II cement. This study aims to present the application of UCS prediction models by machine-learning techniques such as multiple linear regression (MLR), artificial neural network (ANN), and support vector regression (SVR). A total of 84 UCS testing data of cement-stabilised high plastic clay under varying soil–cement ratios (s/c), water-cement ratios (w/c), and curing times are collected from the literature. Test results indicate that the UCS decreases with the increase in s/c and w/c while increasing monotonically with curing time. The model performance is evaluated based on statistical matrices such as RMSE, MAE, and R2. According to the testing results, the SVR model predicted the UCS of PBFC-stabilised marine clay well with the lowest RMSE (24.69 kPa) and MAE (18.36 kPa) and highest R2 (0.984). In contrast, the MLR model showed a greater scatter in the testing data, which is the reason for its poor statistical indices (RMSE = 94.79 kPa, MAE = 76.92 kPa, and R2 = 0.76). The test results finally emphasis the advantage of machine-learning models for in situ ground improvements, which serve as preliminary guidelines towards adopting the appropriate mix ratios.