Hybrid numerical and machine learning investigation of Cs₂PtI₆ lead-free perovskite solar cells for improved PV effectiveness

Lead-free perovskite solar cells (PSCs) are widely recognized as environmentally friendly substitutes for toxic lead-based photovoltaics (PVs). Among them, Cs₂PtI₆ is a promising absorber material owing to its feasible bandgap, strong optical absorption, and well-structured stability. However, overall effectiveness is still limited by recombination among interfaces, non-ideal charge transport, and resistive losses. In this investigation, a hybrid SCAPS-1D and machine learning (ML) approach is developed to optimize Cs₂PtI₆-based PSCs. Systematic simulations are carried out to evaluate electron transport layers (ETLs) (SnO₂, TiO₂, ZnO, CeO₂), hole transport layers (HTLs) (Cu₂O, NiO, PEDOT:PSS, Sb₂Se₃), and back contact metals (Au, Ag, Fe, Mo), where SnO₂ (ETL), Cu₂O (HTL), and Au (back contact) are identified as the most favorable selections under simulation conditions due to superior band alignment and reduced contact resistance. Parasitic parameters are also optimized, where series resistance (Rₛ = 0 Ω·cm²) and shunt resistance (Rₛₕ = 1 × 10⁷ Ω·cm²) yield maximum device performance. The operating condition is optimized at T = 300 K. Under these conditions, the improved device design FTO/SnO₂/Cs₂PtI₆/Cu₂O/Au results in PCE of 26.33%, with V OC = 1.3231 V, J SC = 22.494 mA cm⁻², and FF = 88.46% under standard AM1.5 G illumination. Furthermore, a dataset of 625 SCAPS-generated samples is used to develop ML predictive frameworks for effectiveness analysis. Among them, the Random Forest (RF) model provides the most notable precision with R² = 0.9257 and RMSE = 0.5526. This study provides a reliable strategy to accelerate the structure of effective lead-free PSCs and supports prospective experimental validation and fabrication.