In-Between Randomization Assisted Machine Learning Performance Analysis for Naturally-Sensitized Solar Cells

The utilization of free, renewable, and available solar energy became a focus research area in recent years. Sustainable natural photosensitizers in dye-sensitized solar cells (DSSCs) are among the hot researched topics in the scientific community. Herein, various naturally-sensitized-photoanode-based DSSCs were studied via statistical and machine learning analysis to investigate the possibility to achieve relatively high PCEs in naturally-sensitized DSSCs. Studied photosensitizer (dye) characteristics included chemical structure and bandgap which were correlated to the literature reported PCEs. Input parameters used in models classification training were: the number of π-bonds (PI), the number of anchoring groups (X), HOMO(H)-LUMO(L), and bandgap energy (BG), with only 2 responses regarding the statistical possibility to achieve high PCEs (Yes/No). Both training/testing (80/20)% datasets were carefully chosen to identify the dye controlling parameters responsible for increasing PCEs. The built trained classification models (decision trees) were tested and showed high prediction accuracy. The idea here is to check whether a certain dye and its correlated PCE would achieve below or above the average PCE. Thus, this allowed us to classify the problem according to the selected parameters so that the dyes can be correlated to their BGs and the other parameters. This work shows the potential of applying statistical analysis to natural sensitizers for enhanced charge injection (current density) for renewable, cost-effective, and sustainable energy production.