Machine learning has been extensively used for analyzing spectral data in food quality management. However, collecting high-quality spectral data from miniature spectrometers outside the laboratory is challenging due to various factors such as distortions, noise, high dimensionality, and collinearity. This paper presents an in-depth analysis of food datasets collected from miniature spectrometers to evaluate the data quality and characteristics, by focusing on a case study of olive oil quality check, where various machine learning models were applied to differentiate pure and adulterated olive oil. Furthermore, the impact of pre-processing techniques on data distortions was studied. It presents a comprehensive pipeline, including data pre-processing, dimension reduction, classification, and regression analysis, and deploys different algorithms for comparative classification and regression analysis. The model performances were assessed using 2 separate methods: tenfold cross-validation on an entire dataset with 10% random testing, and an entire test set collected in different environments (multi-session validation). The first validation approach reached classification rates of up to 96.73%, while the second achieved 83.32%. These results demonstrate that cost-effective miniature spectrometers augmented with a suitable machine learning pipeline could execute classification tasks on par with non-portable and more expensive spectrometers. Furthermore, the study highlights the requirement of specialized algorithms to handle different ambient conditions affecting data acquisition and to eliminate performance gaps, making miniature spectrometers suitable for in situ scenarios. This work extends previous research to enable consumers becoming the first line in the defense against food fraud.