AbstractThe unconfined compressive strength (UCS) of soils is essential for both researchers and practitioner geotechnical engineers. UCS is well-understood and standardized for laboratory and field tests. Nevertheless, the large number of environmental and physical governing factors makes the reasonable prediction of UCS complicated. In this paper, a deep learning approach using the multilayer perceptron regressor (MLP) method along with the genetic expression programming (GEP) are used to assess nine variables that contribute to form a reflective multivariate formulation of the UCS. These variables include clay mineral percent (CF), specific gravity (Gs), dry unit weight (γd), saturated unit weight (γsat), natural unit weight (γt), moisture content (MC), void ratio (e), degree of saturation (S), and porosity (n). MLP and GEP are implemented to classify, correlate, rank, and reduce the number of variables that govern the UCS through the application of classification algorithms, importance analysis, interrelations and interdependency analysis of the variables, and functional indicators that shape the UCS. The changes of UCS in line with the variations of void ratio are analytically formulated according to both the critical state soil mechanics and the inverse proportionality between the voided area and soil strength. Moreover, the validity of the UCS principle is examined as opposed to the variation of the clay percent in the soil. Findings show that there shall be a breakpoint (clay percentage) after which the concept of UCS is radically changed due to the presence of a significant amount of frictional and drainable materials in the soil. The breakpoint appears to be centralized between 40% and 55%. The study is concluded by identifying the fundamental soil parameters, providing practical models to evaluate UCS, developing fundamental relationships between UCS and void ratio, and defining the breakpoint of clay content.