Traditionally California bearing ratio (CBR) is obtained by conducting laboratory testing, which is often time-consuming, laborious, and costly. This delays the design and construction processes of important structures. Recently, several researchers have predicted CBR using ML algorithms. This study focused on understanding the uses of various ML algorithms in the prediction of CBR of treated and natural soils, and other applications. Factors like OMC (30%), MDD (29%), LL (25%), PL (20%), and PI (19%) were mostly used as contributing factors for estimating CBR. ANN, RF, and CNN were the best models for predicting settlement of shallow foundation, bearing capacity of piles and slope stability, and landslide identification, respectively. DNN, GEP, and ELM-CSO were the best models in estimating CBR for granular soil, fine-grained soil, and lateritic soil, respectively, and RFR, AB-DT, LR, and ANN for other types of soils. ANN and BBO-MLP were the best models for expansive clay soil treated with HARHA, and pond ash treated with lime and lime sludge, whereas ANN was for lateritic soil treated with cement and RHA, sand with quartz, feldspar, calcite, corund, amorphous, and clay with pozzolan and lime powder, respectively. The quality and quantity of available training data were fundamental to observing the capacity of models, highlighting the importance of richer, better-labeled datasets.
