Performance Low-Cost Spectrometer for Classification of Tempeh Maturity Quality

Tempeh is a traditional Indonesian food popularly consumed for its high nutritional value. Determining the quality of tempeh is an important step in improving quality control of tempeh products. The aim of this research is to develop a low-cost spectrometer using multispectral sensor AS7265x to assess the quality of tempeh maturity over a period ranging from 1 to 5 days. The system comprises the AS7265x sensor chipset, a light source, an Arduino microcontroller, and a personal computer. The sensor covers 18 wavelengths in the 410 to 940 nm spectrum. The results showed that the low-cost spectrometer functioned properly and could read light reflectance. By combining SVM models and optimizing feature selection using filter-based and metaheuristic feature selection algorithms, we were able to classify tempeh with high accuracy. After using wavelengths selection based on mutual information, the accuracy of tempeh can reach 95%, compared to 90% accuracy when not using feature selection. From the accuracy performance results and evaluation metrics obtained, it can be concluded that the combination of SVM-based machine learning with low-cost spectrometer can be effectively and economically used to assess the quality of tempeh maturity.