Probabilistic analysis of simply supported concrete beam using machine learning techniques: a comparative study

Large deflection of a simply supported beam (SSB) carrying central point load is a well-known area in mechanics that has been researched extensively by numerous scholars. Various techniques, including the finite element approach and analytical exact solutions, have been implemented. Few researchers also used machine learning (ML) technique to predict deflection of SSB. This study offers ML based prediction method for deflection of SSB carrying point load and compares the applicability and adaptability of ridge regression model, artificial neural network and Gaussian process regression (GPR) in the reliability investigation of SSB. The central point load (W), beam length (L), beam width (B), beam depth (D), and concrete’s modulus of elasticity (E) are the five key input parameters that these three machine learning models take into consideration when applying them to 200 datasets in order to predict the beam deflection (∆) at the centre of the beam. The effectiveness of the well-established ML models is evaluated using a variety of performance metrics, such as coefficient of determination (R2), variance account factor, A-10 index, root mean square error (RMSE), mean absolute error and mean absolute deviation. Based on performance parameters, the findings indicate that, out of the three suggested machine learning models, GPR had the best predictive performance. This was attributed to its maximum R2 = 1.000 and the lowest RMSE = 1.157E-06 during the training phase, as well as R2 = 0.999 and RMSE = 0.006 during the testing phase. Rank analysis, reliability analysis, regression plot, William’s plot, and error matrix plot are further tools used to assess the performance of the model. First-order second moment (FOSM) approach is used to determine the reliability index (β) of the model and compared with the actual value. A sensitivity analysis is also carried out to determine the impact of each input parameter on the result. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024.