The metal halide perovskite solar cells (PSCs) have been among the most popular research topics in recent years; the power conversion efficiency rose to 23.3% in a few years making this potentially low cost device a serious alternative for the current solar technologies [1]. However, the number of publications increased dramatically and chaotically in a few years; hence machine learning approaches could be beneficial to extract knowledge from such a large and complex accumulation in the literature. Association rule mining method is one of the most common data mining method to find out the relations, frequent patterns and associations in data which cannot be determined by naked eye.

In this work, the literature on perovskite solar cells was systematically reviewed, and a database containing 1921 data points extracted from 800 publications was created using the research articles published between 2013 and 2018 (until May 31 of 2018). This database was assumed to represent the literature fairly well considering that it covered about 10% of the Web of Science articles. Then, this database was analyzed using association rule mining to capture the major trends and frequent patterns in the collection of works.

The data analyzed using simple descriptive statistics first; the performance evolution with time as well as the average and distribution of PCE values of different materials and deposition methods were investigated. For example, the average efficiencies obtained with MAPbI₃ and MAPbI_3-xCl_x were found to be nearly same while FA based and mixed cation cells gave higher efficiency. Additionally, the potential of some rarely used materials (like Cs containing triple cation perovskites), inorganic HTLs and PTAA as HTL alternative for both regular and inverted cells were also emerged in this simple analysis.

Then the association rule mining analysis was employed to detect the most frequent items appeared in dataset for high performance. The factors such as mixed cation perovskites, DMF+DMSO as solvent, chlorobenzene as anti-solvent and two or three times spinning as the one-step coating technique emerged as the effective ways of obtaining cells with PCE higher than 18.0%. Similarly, relatively less frequently used factors like LiTFSI+TBP+FK209 as HTL additive and SnO₂ as ETL layer were also detected as the alternatives for high performance.