State of the art food quality measurement techniques fail to determine quality of the food item once it is packed and sealed in an enclosed package. The disclosure herein generally relates to food quality prediction, and, more particularly, to a system and method for predicting liquid food quality in a non-invasive manner. A near infra-red (NIR) radiation is transmitted through a semi-transparent opening configured on an enclosed package containing a liquid food item and the resulting NIR reflection spectra is collected. The quality of the liquid food item is estimated by correlating a plurality of features derived from the NIR reflection spectra with the concentration of the biomarker contained in the liquid food item, using a trained machine learning model and the remaining shelf life of the liquid food item is estimated based on the concentration of the biomarker.