Assessment of Uniaxial Strength of Rocks: A Critical Comparison Between Evolutionary and Swarm Optimized Relevance Vector Machine Models

The present study compares the evolutionary and swarm-optimized relevance vector machine (RVM) models to find the optimal model for computing rocks’ uniaxial compressive strength (UCS_R). This study will help rock engineers compute the UCS_R and escape from the arduous laboratory procedures. To aim this, the polynomial, sigmoid, Laplacian, linear, exponential, and Gaussian functions have been used to create six RVM models, further hybridized by each genetic (GA) and particle swarm optimization (PSO) algorithm. The capabilities of models have been measured and analyzed using 131 data points. Furthermore, the multicollinearity analysis is performed to investigate the effect of database multicollinearity. The model comparison reveals that the GA-optimized laplacian kernel-based (Lap_GA) RVM model has assessed UCS_R with excellent performance metrics, i.e., a20 index = 62.96, performance index = 1.6632, agreement index = 0.8309, root mean square error = 25.5219 MPa, and correlation = 0.9289, better than other RVM models. Also, model Lap_GA achieved first rank with a 202 score, including both learning phases. The cost computation and generalizability results also show the superiority of the Lap_GA model in assessing UCS_R.