Eggsperimenting

A data scientist applies computer vision and logistic regression to predict chicken egg hatchability in a personal homesteading project. Using a Kaggle dataset of 4,000+ candled egg images, a CNN was trained achieving 96% accuracy and 0.91 AUC. The model assigned fertility probabilities to 18 named eggs across incubation days, which were combined with physical measurements (shell porosity, weight loss, cleanliness) into a composite viability score. Eggs were grouped into narrative categories: Minor Characters, Cliffhangers, and Protagonists. Of 18 eggs, 10 hatched. The logistic regression model showed strong alignment with true outcomes, with no false negatives, though small sample size limited generalizability. The experiment highlights how deep learning, quantitative analysis, and human judgment each contribute uniquely to prediction tasks.