Artificial intelligence for honey integrity in Ghana: A feasibility study on the use of smartphone images coupled with multivariate algorithms

Abstract

Rapid, onsite, and simple detection of honey integrity in real time provides consumers with the assurance that the honey purchased is free from adulteration such as sugar syrup. In this study, smartphone based camera technique was used to differentiate authentic honey from adulterated ones (using authentic honey samples from different locations and adulterated samples made by spiking authentic ones with sugar syrup in the laboratory) by employing two supervised machine learning algorithms. Experimental results showed that the prediction model based on Random Forest (RF) was superior to K-nearest neighbour (KNN). The optimum results were assessed based on the prediction rate, specificity, and sensitivity. The performance of the RF model was 100% accuracy while the specificity of 100%, and sensitivity of 100%. These findings could be exploited for reliable and rapid classification of honey integrity in Ghana and West Africa in general. This will further improve consumers’ confidence in the honey trade due to the ease and availability of smartphone technology.