Prediction of milk varieties and milk quality parameters with ensemble learning methods

Milk is one of the primary foods consumed by living beings for growth and development. Due to its high nutritional value, being a source of protein, calcium, and vitamins, it is an essential natural food recommended for all age groups. In addition to its high nutritional value, milk is also a fundamental food that regulates bodily functions, supports body development, and plays an important role in bone and tooth formation. For this reason, milk and dairy products are consumed by almost everyone in society. Due to its wide consumption, it is possible to encounter misleading practices such as adulteration, dilution, etc. Determining the quality of milk is of great importance against such threats. In this thesis study, two different datasets were worked on. In the first dataset, created by using values obtained from measurements taken with the help of an FTIR device on different types of milk samples (cow milk, sheep milk, goat milk, commercial milk, and milk powder), milk types were attempted to be predicted using classification methods. In the second dataset, published on the Kaggle open-source platform, milk quality was attempted to be predicted using classification methods. Traditional machine learning methods such as KNN and Decision Trees, as well as ensemble learning methods such as Random Forest, Gradient Boosting, AdaBoost, and Voting, were utilized as classification methods. The prediction performance of the models was evaluated with four performance metrics: accuracy, sensitivity, precision, and F-score. Looking at the results, in predicting milk types with the first dataset, KNN, Decision Trees, Random Forest, Gradient Boosting, and Voting methods showed perfect (1.00) performance, while the AdaBoost method showed lower (0.8) performance. In predicting milk quality with the second dataset, Voting, Gradient Boosting, Decision Trees and Random Forest (0.99) methods showed the highest performance. When the performances of the used methods are evaluated, the high overall success shows that these traditional machine learning and ensemble learning methods can be used in studies in this field. KEYWORDS: Classification, milk, ensemble learning, FTIR Spectroscopy, adulteration