The abundant quantities of impurities resulting from the production of phosphoric acid by dihydrate process promotes fouling in the heat exchanger which is manifested by quick clogging of the tubes, outlet acid temperatures rise to values close to safety measures and operating cycles less than five days. During the acid concentration operation, fouling leads to a significant drop in the overall heat transfer coefficient, which is highly dependent on the thermal efficiency of the heat exchanger. This requires an instantaneous follow-up. The present work aimed at modeling thermal efficiency in a cross-flow heat exchanger by the collection and treatment of the operating data of the concentration loop. To this end, Principal Component Analysis (PCA) was selected for dimensionality reduction. A first fouling factor modeling by Projection to Latent Structures (PLS) showed a decent estimation of the deposition phenomenon with a correlation coefficient R-2 and prediction ability equal to 0.925 and 76%, respectively. In order to improve the results obtained by PLS, Artificial Neural Network (ANN) with backpropagation method was used thereafter. Among the 75 trained topologies, the best performance was obtained with a network consisting of one hidden layer with 7 neurons using tangent sigmoid transfer function for the hidden and output layers. The optimized ANN model for thermal efficiency prediction resulted in reliable quality indices which reflect that the model accurately tracks the variability within measurements with AARD = 0.0639%, MSE = 0.00003, RMSE = 0.00573 and r(All)(2) = 0.9998. (C) 2019 Elsevier Ltd. All rights reserved.