DEVELOPMENT OF VALUE CONVERSION USING MACHINE LEARNING ALGORITHM FOR MILK FAT DETECTION DEVICE USING NIR-SPECTROSCOPY METHOD

• Main Author: Prima Dewi, Elzania
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/54365
• Institution: Institut Teknologi Bandung

Milk contains a lot of calcium so it is important for bone health. Milk also contains more than other vital nutrients including protein, carbohydrates, vitamins, minerals, and fats. The U.S Department of Agriculture (USDA) recommends about three cups of milk (732 ml) a day for adults and children age 9 and older to help meet daily dietary needs for nutrients. Based on the fat content, milk is divided into three types, namely pure milk (about 3.5% of fat), low fat milk (2% of fat), and skim milk (0% of fat). Lowfat milk and skim milk can be used as alternatives to a healthy and balanced diet. In this final project, research is designed a device to detect the percentage of fat in milk using Near-Infrared Spectroscopy (NIRS) method. The device can display the value of the voltage (V) of the photodiode that captures the light from the LED that passes through a cuvette filled with liquid milk. The resulting voltage value will be one of the input features for the conversion of values to milk fat percent using the Machine Learning (ML) algorithm. The implementation of ML in this final project uses the Python programming language. Python is the main programming language that is widely used in ML and Artificial Intelligence (AI). In continuous data processing, ML will use various libraries from Python, such as Scikit-learn, Pandas, Keras, Matplotlib, and TensorFlow. The result of conversion of ML values with Neural Network (NN) results in better accuracy than linear regression, especially for the case of a linear regression that has more than one independent variable. In the implementation of this NN model, the smallest Mean Absolute Error (MAE) value is 0.1318.