The development of an algorithm for fermented milk microbial quality assessment

Fermented milk is widely produced and consumed in Cameroon; however, small-scale producers rarely perform microbiological testing due to high laboratory costs, increasing the risk of distributing contaminated products. This study developed and validated a rule-based decision-tree algorithm integrating sensory attributes, acidity, and Gram-staining results for rapid classification of fermented milk quality. Fifty samples collected between July and September 2022 from five quarters in Yaoundé were analyzed for sensory characteristics, pH, titratable acidity, microbial loads, and Gram-staining profiles. Relationships among these parameters were assessed using ANOVA. Sensory evaluation indicated generally acceptable color and taste across all markets, while aroma and texture varied significantly (p < 0.05). Physicochemical analysis revealed variability in fermentation, with pH ranging from 3.51 to 4.11 and titratable acidity from 0.95% to 1.53%. Coliform counts ranged from 1.08 to 3.86 log₁₀ CFU/mL. Staphylococcus species were detected only in traditional products. Acidity and Gram-staining patterns were strongly associated with microbial safety (r =-0.56 to r=-051). The decision-tree model based on Gram stain, aroma, texture, and acidity achieved 72% accuracy, with high recall (0.93) for contaminated samples. Pruning improved results, while Random Forest was best, with 84% accuracy, 0.82 precision, and 0.94 recall. Gram-stain morphology was influential. Cross-validation confirmed robustness (mean accuracy 89.6%, 95% CI 0.852–0.940). These findings reveal variability in milk quality in Yaoundé and demonstrate machine learning’s potential as a rapid, low-cost screening tool for food safety. Future should include validation and larger datasets to improve performance.