Abstract
<jats:p>Estimation of raw cow’s milk spoilage timelines is critical for food safety, yet traditional assays remain costly and destructive. This paper introduces a non-invasive alternative integrating a Microscopic Image Acquisition Setup with a two-stage hybrid machine learning framework. The methodology extracts 𝐿∗𝑎∗𝑏∗ color parameters from digital RGB images to track milk degradation at am- bient temperatures. Validated on a balanced dataset of 4,524 data points via a 70/30 split, the pipeline first deploys an optimized 𝑘-NN classifier (𝑘 = 9) to identify discrete degradation phases with 99.93% accuracy (error rate: 0.0007). This prediction subse- quently drives a multivariable linear regression model to map continuous temporal kinetics. The regression model accounts for 77.22% of the spoilage variance (𝑅2 = 0.7722) with an MAE of 18.25 min . Within our window, this MAE induces a bounded 10% prediction error, successfully capturing the non-linear, exponential microbial degradation kinetics without overfitting. This lightweight architecture enables instantaneous inline screening in decentralized dairy collection centers.</jats:p>