Perovskite solar cells (PSCs) have attracted significant attention in recent years due to their exceptional power conversion efficiencies and low-cost fabrication potential. However, accurately modeling their J–V characteristics under varying irradiance conditions remains challenging, as conventional experimental methods require considerable time, cost, and experimental effort. In this work, a machine learning-based approach was developed to overcome these limitations by training a Multi-Layer Perceptron (MLP) artificial neural network capable of predicting PSC performance with high precision. The model was trained on a large-scale, simulation-generated dataset covering diverse irradiance levels, using irradiance intensity and voltage as inputs and current as the output. The Levenberg-Marquardt algorithm enabled fast convergence and low prediction error. The proposed Artificial Neural Network (ANN) achieved correlation coefficients above 0.9996 and very low Mean Squared Error (MSE) values across training, validation, and testing. Comparative analysis showed a close match between the predicted J–V curves and the simulation data, confirming the model’s reliability. These results indicate that the proposed ANN offers a cost-effective and scalable solution for PSC performance modeling, potentially accelerating the optimization and deployment of next-generation photovoltaic technologies.