Modelling the anthocyanin content in black mulberry using colour parameters

The pH differential method is commonly used to determine the total monomeric anthocyanin content of fruits. This method requires intensive labour and time as well as expensive chemicals. The aim of this thesis is to develop a simpler and faster model for the determination of anthocyanin content of black mulberry fruit, which will be an alternative to the existing method. Within the scope of the thesis, a black mulberry (Morus nigra) K60 genotype in Tokat Gaziosmanpaşa University Agricultural Application and Research Centre was used. The fruits were harvested at different stages of ripeness in July 2022 and 2023. Firstly, external colour (L*, a*, b*) measurements were made and then pH, soluble solid content (SSC) and anthocyanin analyses were performed. The data obtained were divided into two groups as independent (L*, a*, b*, pH, SÇKM) and dependent (anthocyanin) variables. Four different machine learning algorithms were utilised for anthocyanin prediction. These are Multilayer Perceptron, SVM, KNN and RandomForest. Correlation coefficient (r), Mean absolute error (MAE), Root Mean Square Error (RMSE), Relative Absolute Error (RAE) and Root relative squared error (RRSE) were used as performance evaluation criteria of the models. As a result of the study, among the machine learning algorithms used, the RandomForest algorithm achieved the most successful results in predicting anthocyanin values. RandomForest was followed by the KNN, SVM, and MultilayerPerceptron algorithms, respectively. The performance metrics for the RandomForest algorithm were determined as follows: correlation coefficient (r) 0.8498, MAE 52.779, RMSE 79.0309, RAE 43.3826 %, and RRSE 52.7319 %.