Analysis of Ruddlesden-Popper and Dion-Jacobson 2D Lead Halide Perovskites Through Integrated Experimental and Computational Analysis

Two-dimensional (2D) lead halide perovskites (LHPs) have captured a range of interest for the advancement of state-of-the-art optoelectronic devices, highly efficient solar cells, next-generation energy harvesting technologies owing to their hydrophobic nature, layered configuration, and remarkable chemical/environmental stabilities. These 2D LHPs have been categorized into the Dion-Jacobson (DJ) and Ruddlesden-Popper (RP) systems based on their layered configuration respectively. To efficiently classify the RP and DJ phases synthetically and reduce reliance on trial/error method, machine learning (ML) techniques needs to develop. Herein, this work effectively identifies RP and DJ phases of 2D LHPs by implementing various ML models. ML models were trained on 264 experimental data set using 10-fold stratified cross-validation, hyperparameter optimization with Optuna, and Shapley Additive Explanations (SHAP) were employed. The stacking classifier efficiently classified RP and DJ phases, demonstrating a minimal variation between the sensitivity and specificity and achieved a high Balance Accuracy (BA) of (0.83) on independent test data set. Our best model tested on 17 hybrid 2D LHPs and three experimental synthesized 2D LHPs aligns well experimental outcomes, a significant advance in cutting edge ML models. Thus, this proposed study has unlocked a new route toward the rational classification of RP and DJ phases of 2D LHPs.