Jordan Journal of Civil Engineering

Influence of Database Size on Artificial Neural Network Results for the Prediction of Compressive Strength of Concretes Containing Reclaimed Asphalt Pavement


Rim Larbi; El Hadi Benyoussef; Meriem Morsli; Mahmoud Bensaibi; Abderrahim Bali;


The main objective of this study is to show the influence of database size on a considered architecture of multilayer feed forward ANN results for predicting the compressive strength of concretes containing reclaimed asphalt pavement (RAP). On the basis of factorial design, polynomial models were developed for each data series reported in the literature as well as our own experimental results, in order to generate data with different increment ratios (1, 0.5, 0.2 and 0.1). The database passed from 104 data to 130, 336, 1530 and 5440 data, respectively. Training and testing the model showed the efficiency of using ANN models to predict the compressive strength of RAP concretes. The more the database size is increased, the more the results are improved. Better results were obtained when data was generated with an increment ratio of 0.5. The proposed approach proves that factorial design can be used to generate data when needed.


Compressive strength, Reclaimed asphalt pavement concrete, Artificial neural networks, Database size, Data generation, Factorial design.