Salp Swarm Algorithm-optimized Support Vector Machines For Lateral Strength Prediction of RC Columns
The existing methods of predicting the lateral strength of RC columns lack generalization performance,where methods of predicting the flexural strength of ductile columns cannot be used to predict the shear strength of non-ductile columns,and vice versa.While current machine learning methods can solve this problem,they cannot au-tomatically remove a large number of redundant and irrelevant features from the dataset,which in turn increases the complexity of the ML model and leads to overfitting.To this end,this paper proposes a new method called salp swarm algorithm-optimized least squares support vector machines(SSALS-SVM),which can remedy the afore-mentioned problems.Based on a given data set,SSALS-SVM can adopt the salp swarm algorithm(SSA)to au-tomatically eliminate redundant and irrelevant features and select the most representative feature subset with weak correlation among features to form an optimal feature combination,while the hyperparameters governing the nonlin-ear fitting ability of LS-SVM are also optimized.In this way,the optimized prediction model can not only identify the design variables that influence the lateral strength of ductile and non-ductile columns,but also reflect the nonlinear mapping relationship between the optimal feature combination and lateral strength of RC columns.The generalization performance of proposed SSALS-SVM for predicting the lateral strength of RC columns is verified by comparing with existing prediction models based on 248 experimental data of RC columns.Numerical results show that the generalization performance of proposed SSALS-SVM can be enhanced up to 83％higher than that of exist-ing prediction models.
万方数据
维普
