Tabriz is located in one of the important seismic areas of the world and has witnessed severe earthquakes in the past centuries. Earthquake is associated with multiple risks including geotechnical risks which affected many cities around the world. One of these important risks is the phenomenon of soil liquefaction. Soil liquefaction is the reason for many damages caused by earthquakes which can cause lots of damage to vital arteries of cities, mines, pipe lines and the buried structures in the soil. One of the recent challenges in dealing with liquefaction is utilizing intelligent tools for predicting the effects of this phenomenon in soil layers. For this purpose, a total number of 100 soil samples are collected, while an empirical approach is also developed for achieving Liquefaction Potential Index (LPI) by means of the depth of the soil layers, SPT values, penetration indices, fines content percentages, ground acceleration, and water level of the soil samples. For prediction purpose, the recently developed configuration of the Gradient Boosting (GB) methods is utilized as the main approach while the Artificial Neural Network (ANN) and the Decision Tree (DT) approaches are utilized for comparative investigations. For validation process, 10% of the samples are utilized in a stochastic way to intelligently evaluate the capability of the GB method in contrast to the alternative approaches. The results demonstrate the capability of the GB approach in providing efficient predictive results in dealing with the LPI prediction problem. Regarding the training phase, GB provided the maximum absolute error of 3.44 x 10-8 while the DT's result is partially competitive with maximum absolute of 3.15. Based on the test phase, GB can provide the lowest Mean Squared Error (MSE) of 0.09 while the DT with 0.11 and ANN with 3.25 have the other ranks. The GB is capable of reaching to lowest Mean Absolute Percentage Error (MAPE) of 3.64 in this phase while the DT with 3.07 and DT and ANN with 4.97 and 26.05 have second and third ranks respectively. 0.98 with 2% inaccuracy rate.
来源平台：BULLETIN OF ENGINEERING GEOLOGY AND THE ENVIRONMENT
