Hybrid machine learning models for predicting compressive strength of self-compacting concrete: an integration of ANFIS and Metaheuristic algorithm

Self-compacting concrete (SCC) has become increasingly popular due to its superior workability, segregation resistance, and compressive strength. As the traditional methods for strength prediction are costly and time-intensive, this study explores machine learning (ML) techniques as efficient alternatives for SCC compressive strength prediction. Three state-of-the-art hybrid Adaptive Neuro-Fuzzy Inference System (ANFIS) models, optimised using Firefly Algorithm (FA), Particle Swarm Optimization (PSO) and Genetic Algorithm (GA). For the purpose, a robust dataset of 366 instances and 7 input parameters is taken from the literature. After data analysis and pre-processing, the hyperparameters of the models are tuned for the best-fit model and tested on unforeseen data. ANFIS-FF model stands out as the best-performing model (RTR2 = 0.945 and RTS2 = 0.9395) in both training and testing phases, closely followed by ANFIS-GA. All the hybrid models outperform the traditional ANFIS model, outlining the significance of hybridisation, however, ANFIS-PSO lags behind the other two hybrid models. The study highlights the importance of integrating machine learning with metaheuristic algorithms for tackling complex engineering problems like design of optimal SCC mix design, minimising material waste and ensuring cost-effectiveness. It serves as a benchmark for future research comparing hybridisation techniques and serves as a starting point for integrating other algorithms with ANFIS.