Neural network and decision tree-based machine learning tools to analyse the anion-responsive behaviours of emissive Ru(ii)–terpyridine complexes

Abstract

We implemented both neural network and decision tree-based machine learning tools to analyse the anion-responsive behaviours of two heteroleptic Ru(II) complexes based on two tridentate ligands, 2,6-bis(benzimidazole-2-yl)pyridine (H₂pbbzim) and substituted terpyridine ligands, tpy-Ar with Ar = 2-naphthyl and 9-anthryl groups. The secondary coordination sphere of the complexes is decorated with two imidazole NH moieties, benefitting from the anion sensing characteristics of the complexes previously reported by us. Considerable change in their absorption, emission as well as electrochemical and spectroelectrochemical responses occur in the presence of selected anions. Restoration of their initial states is made possible by acid and the process is reversible. We utilized their spectral, electrochemical and spectroelectrochemical responses upon the influence of anions and acid to mimic the operations of YES-NOT and set-reset flip-flop logic gates. We also implemented machine learning tools such as artificial neural networks (ANNs), adaptive neuro-fuzzy inference system (ANFIS) and decision tree (DT) regression to analyse and forecast the experimental data and can thus reduce the time and expenditure associated with the execution of comprehensive sensing experiments. The outcomes of the ANN, ANFIS and DT methods were also tallied with the experimental results. Among the three models, the outcomes derived from DT regression analysis turned out to be excellent with almost zero error. Thus, the applied machine learning based tools could be regarded as a prospective anion-responsive data model for the studied complexes.