The current-voltage (I–V) characteristics of perovskite solar cells (PSCs) exhibit significant dependence on irradiance, which is challenging to capture fully using traditional modeling approaches. This study presents a comparative performance analysis of five distinct machine learning (ML) techniques-Linear Regression (LR), Support Vector Machine (SVM), Generalized Additive Model (GAM), Gaussian Kernel Regression (GKR) and Gaussian Process Regression (GPR)-for modeling the irradiance-dependent I–V curves of PSCs. The models were trained and tested using a large-scale dataset derived from drift diffusion (DD) simulations, encompassing I–V characteristics across five irradiance levels ranging from 10 to 100 mW/cm². Model performance was evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE) metrics. Results demonstrate that GAM consistently achieved the highest predictive accuracy, yielding the lowest error scores on both the training (80 %) and testing (20 %) datasets (RMSE: 0.0956, MSE: 0.0091, MAE: 0.0313). Although GPR exhibited comparable performance and approached the accuracy of GAM, it still produced slightly higher error values. In contrast, LR and SVM showed systematic errors in nonlinear regions, while GKR delivered moderate performance. These findings highlight GAM as a superior tool for modeling the irradiance-dependent electrical behavior of PSCs, offering high accuracy, interpretability, and data efficiency. This work supports the practical application of ML-based surrogate models to reduce experimental burden and accurately predict PSC performance under diverse illumination conditions.