In this paper, we report an intelligent model based on ANN to optimize the performance of an internally cooled membrane-based liquid desiccant dehumidifier (IMLDD). IMLDD can effectively mitigate dehumidification deterioration caused by changes in the temperature of the desiccant solution. The mediums of desiccant solution and air are isolated by means of a semi-permeable membrane on both sides in the IMLDD. The temperature of the desiccant solution is reduced by the cooling media that flows through the tubes placed within the solution channels. Generally, many fluid flow parameters like air, cooling water, desiccant solution, etc., play a critical role in controlling the performance of an IMLDD. For our study, inlet air temperature (T-ai), inlet concentration of the desiccant solution (C-dsi), flow rate of the desiccant solution at the inlet (<(m)over dot>(dsi)), and inlet cooling temperature of water (T-cwi) have been considered as the operating parameters/conditions. The outputs or responses namely dehumidification efficiency (eta(dh)), Exergy efficiency (eta(ex)), and unmatched coefficient (xi(um)) analyze the performance of the IMLDD. The data comprising of massive input-output was achieved using the response surface methodology (RSM) based central composite design (CCD). Back propagation algorithm (BP), artificial bee colony (ABC), and genetic algorithm (GA) models were used to train the neural network (NN) parameters using the data collected from the CCD based response equation. Forward and reverse mapping models were developed using the trained ANNs. Forward modeling predicts the performance parameters of the IMLDD (i.e., eta(dh), eta(ex), and xi(uc)) for known combinations of operating parameters (i.e., T-ai, C-dsi, <(m)over dot>(ds)(i), T-cwi). Similarly, reverse modeling aims at predicting the operating conditions for a known set of performance parameters. The performances of the employed NN models were tested using fifteen arbitrarily generated test cases. The experimental and neural network predicted results were found to be in line with each other for both forward and reverse models. The forward modeling results could assist engineers with off-line tracking, by predicting the response without executing experiments. The reverse modeling prediction will aid in dynamically adjusting the operating parameters to achieve the optimal thermodynamic output characteristics.