Predictive analytics for tea leaf aging and quality degradation

Abstract

Tea quality directly influences flavor, marketability, and economic value. Traditional approaches to assessing tea leaf quality rely on manual inspection and cannot anticipate future degradation, leading to post-harvest losses. This paper introduces a two-stage, microservices-based predictive analytics system that empowers tea producers with forward-looking insights. In the first stage, a state-of-the-art object detection model processes harvested leaf images to classify quality into four tiers. In the second stage, a Random Forest classifier forecasts daily quality degradation over a fifteen-day horizon by combining leaf characteristics with environmental data-temperature, humidity, and rainfallfetched from public APIs and efficiently cached to minimize redundant calls. Explainable AI techniques distill each day’s prediction into the top three driving factors, presented in farmer-friendly language alongside actionable harvest recommendations. Deployed as serverless services and accessed through a mobile interface, the framework delivers scalable, low-latency predictions. This research addresses the critical need for proactive quality management in tea production by uniting image-based classification, environmental data integration, and explainability into a novel, end-to-end solution.