Classification of smart films for spoilage detection in food products

Smart films incorporating natural pigments offer a sustainable, non-invasive solution for detecting spoilage in perishable food products. These intelligent indicators undergo visible colour changes in response to volatile nitrogen compounds released during microbial degradation, enabling real-time freshness monitoring. However, current applications rely heavily on subjective human interpretation, limiting their reliability across diverse conditions. Existing applications of machine learning techniques are limited to a small sample set with specific settings of film development and data collection. This study explores a cost-effective alternative to classify spoilage stages using image data and metadata extracted from the published literature. A dataset of images of anthocyanin-based smart films was compiled and standardised, with samples categorised as g'Spoiled' or 'Not Spoiled'. Supervised learning models, including K-Nearest Neighbour, Naive Bayes, Decision Tree, Support Vector Machines, Artificial Neural Networks and Random Forest, were trained to predict spoilage, achieving the promising baseline accuracy.