An Empirical Comparison Of Multiple Linear Regression And Artificial Neural Network For Load Slab-Deflection Prediction Modeling

Abstract

— All structures are subjected to different kinds of environment and conditions that result to
deflection. Reinforced concrete (RC) slabs may exhibit deflections due to different types of loadings in the
form of cracks; excessive deflections pose problems in structural stability and may reduce overall integrity
of the whole structure. In this research, a load-deflection behavior prediction of RC slabs with the use of
artificial neural network (ANN) and multiple linear regression (MLR) is designed. RC slab properties are
used, such as thickness, effective depth, span length; concrete compressive strength, concrete tensile
strength, concrete Young’s modulus; steel yield strength, steel Young’s modulus; ultimate load and
maximum deflection, as input variables for the models, which will then be simulated using MLR analysis
and ANN (dubbed as models A and B, respectively). Results showed that model A have coefficient of
correlation (R) equal to 0.896 and mean square error (MSE) of 68.137 while model B obtained an R and
MSE of 0.998 and 1.259, respectively. Calculating mean absolute percentage error (MAPE) for models A
and B yielded 24.105% and 4.241%, respectively, which means that the ANN model yielded closer
predicted values as compared to the measured deflection values. It can be concluded that the predictions
made with ANN displayed better results than MLR analysis and thus considered as a more efficient way for
this prediction approach.