Investigating the Soil Unconfined Compressive Strength Based on Laser-Induced Breakdown Spectroscopy Emission Intensities and Machine Learning Techniques

Laser-induced breakdown spectroscopy (LIBS) is a remarkableelementalidentification and quantification technique used in multiple sectors,including science, engineering, and medicine. Machine learning techniqueshave recently sparked widespread interest in the development of calibration-freeLIBS due to their ability to generate a defined pattern for complexsystems. In geotechnical engineering, understanding soil mechanicsin relation to the applications is of paramount importance. The knowledgeof soil unconfined compressive strength (UCS) enables engineers toidentify the behaviors of a particular soil and propose effectivesolutions to given geotechnical problems. However, the experimentaltechniques involved in the measurements of soil UCS are incrediblyexpensive and time-consuming. In this work, we develop a pioneeringtechnique to estimate the soil unconfined compressive strength usingartificial intelligent methods based on the spectra obtained fromthe LIBS system. Decision tree regression (DTR) and support vectorregression learners were initially employed, and consequently, theadaptive boosting method was applied to improve the performance ofthe two single learners. The prediction power of the established modelswas determined using the standard performance evaluation metrics suchas the root-mean-square error, CC between the predicted and actualsoil UCS values, mean absolute error, and R (2) score. Our results revealed that the boosted DTR exhibited the highestcoefficient of correlation of 99.52% and an R (2) value of 99.03% during the testing phase. To validate themodels, the UCS values of soils stabilized with lime and cement werepredicted with an optimum degree of accuracy, confirming the models'suitability and generalization strength for soil UCS investigations.