Reliability analysis of pile bearing capacity using state-of-the-art hybrid Neuro-Fuzzy ML approaches

This study presents a new state-of-the-art machine learning-based methodology for reliability analysis of pile foundations utilizing an extensive experimental dataset of 472 static pile load test results. This approach bridges the research gap in the lack of reliability analysis studies using robust experimental data. Most of the studies utilized either conventional models or small datasets. Considering the variable nature of the soil, traditional approaches are not reliable. A probabilistic approach based on the Adaptive Neuro-Fuzzy Inference System (ANFIS), integrated with Particle Swarm Optimization (PSO), Genetic Algorithm (GA), and Firefly Algorithm (FF), to predict the bearing capacity of piles is concluded as a robust methodology. The comparative analysis demonstrated that ANFIS-PSO exhibited superior performance, achieving an R2 of 0.949 in the training phase and 0.953 in the testing phase. Furthermore, model reliability was evaluated using the Reliability Index (β), which indicated that ANFIS-PSO (β = 1.990 for training and β = 1.970 for testing) demonstrated a reliability index closer to that computed from the experimental output values (β = 1.936 for training and β = 1.947 for testing). This study demonstrates that ANFIS-PSO enhances pile capacity prediction and structural safety in deep foundations.