The aim of this study is to optimize the geothermal energy aided absorption refrigeration system using NH3-H2O as the working fluid. A total of 3660 different designs, with different solution fractions and working parameters, were analyzed by means of energy-exergy and net present value (NPV) analysis. The obtained data was modeled by a novel two stage artificial neural network (ANN) with 14650 data points. Of this, 10248 points were used for training, and the remaining used for testing. The best topology of ANN were performed by using the back-propagation learning algorithm with three different variants such as Levenberg-Marquardt (LM), Pola-Ribiere Conjugate Gradient (CGP), and Scaled Conjugate Gradient (SCG). According to ANN results, the error rates were determined in an acceptable range change between 0.07% and 6% for the engineering applications. The R-2 values of best network structures were calculated in higher acceptable range change between 0.9958 and 1.000 for LM. The optimum designs were determined using the obtained weights and biases of the best ANN topology, yielding a coefficient of performance (COP) and exergy efficiency (epsilon) of 0.5722 and 0.6201, respectively. NPV values were respectively calculated as 1.778 million US$, 6.328 million US$ and 27.183 million US$ for quince, apple and grape. (C) 2016 Elsevier Ltd. All rights reserved.