Development of a Learning Based Shele- Life Prediction Model for Kaju Katli

Abstract

Kaju katli is a widely consumed traditional Indian confectionery prepared primarily from cashew nuts and sugar. Despite its popularity, the product has a limited shelf life due to physicochemical and microbiological deterioration during storage, including moisture loss, lipid oxidation, and microbial growth. Conventional shelf-life determination methods rely on repeated laboratory testing, which is time-consuming, labour-intensive, and expensive. This study aims to develop a machine learning-based approach for predicting the shelf life of kaju katli using quality parameters obtained through laboratory analysis. The work focuses on integrating food science with artificial intelligence to provide a rapid, reliable, and cost-effective alternative to conventional shelf-life estimation. The study evaluates key physicochemical and microbiological parameters, namely moisture content, water activity (aw), free fatty acid (FFA), standard plate count (SPC), and yeast and mould count (YMC), as indicators of product quality during storage. Experimental data collected from laboratory analyses are organized and enhanced through feature engineering techniques to improve dataset quality and predictive capability. A Random Forest regression model is developed to establish relationships between these quality attributes and the remaining shelf life of kaju katli. The model is trained and evaluated using standard performance metrics to ensure prediction accuracy and reliability. The proposed machine learning model is expected to accurately predict shelf life by capturing the complex interactions among physicochemical and microbiological parameters that influence product deterioration. The approach minimizes dependence on prolonged storage studies while providing a scientific basis for quality monitoring and decision-making. The findings are expected to demonstrate that moisture content, water activity, FFA, SPC, and YMC are significant predictors of shelf life and can effectively support intelligent food quality assessment. Overall, this research highlights the potential of machine learning in modern food engineering by offering an efficient tool for shelf-life prediction of traditional confectionery products. The developed model can help manufacturers optimize storage conditions, reduce food wastage, improve inventory management, and enhance consumer confidence by providing more accurate shelf-life estimates. Furthermore, the methodology can be extended to other traditional Indian sweets and similar food products, contributing to the advancement of intelligent food processing and quality assurance systems.