Now a days provskite solar cells have been very interesting candidates for commercialization due to their low production cost. However, their weak long-term stability remains as an obstacle of this group of materials.
In this work a database contains of 354 halide provskite candidates including single provskite in the form of ABX and double provskite in the form of AB1B2X have been considered as a training data set. In single provskite “A” indicates K, Rb or Cs, “B” denotes Sn pr Pb and “X” can be Cl, Br or I while in double provskite formula, “A” stands for Na, K, Rb or Cs, “B1” is Na, K, Rb, Cu, Ag, In and Tl, “B2” denotes Sb, Bi, Ca or In and “X” represents a halogen atom such as Cl, Br or I.
While the machine is trained using the above data set, it has been validated using the experimental observation of provskite formability of 246 compound that are not included in the training data set.
There is a very good agreement between the results derived from the machine learning (ML) method and the one from experimental studies which shows the ML method is enough accurate to study the stability of other provskite rather than those 246 experimentally known compound.
This means that the generalization potential of the introduced machine learning model has been selected in a good way. In this work, it has been proven that a machine learning method based on density functional theory data set can be used to predict stable provskite for experimental works.