Determinação de concentração de amido em curcuma longa utilizando machine learning

Abstract

The food and pharmaceutical industries are strictly regulated in order to guarantee the quality of the manufactured and industrialized products. The usual methodology applied in these processes involves chemical procedures that can be wasteful and costly. Thus, the use of computational methods to contribute and optimize the quality process is of great importance. Turmeric (Curcuma longa) is a tuberous root herbaceous plant. In many places around the world, its dried and ground root is used mainly as a condiment, dye, and drug. There are several studies that point out its health benefits, from antioxidant and anti-inflammatory activities to help combat diseases such as cancer, arthritis, Alzheimer's, and cardiovascular diseases. Saffron is among the 7 most adulterated foods in the last 30 years, with starch as its main contaminant. Therefore, this work explores the use of machine learning for the determination of starch concentration in digital images of turmeric samples. The following models were tested: multiple linear regression, decision tree, and support vector machine. The support vector machine model obtained the best results, with a root mean square of 6.52, a mean absolute error of 5.78, and a coefficient of determination of 0.93. It is concluded that support vector machine is a good tool for the determination of starch concentration for turmeric.