Traditional methods for assessing bread edibility rely on subjective visual and sensory analysis, which can be inconsistent and labor-intensive. This study introduces a novel data-centric approach to automate bread edibility classification using an Integrated Feature Set (IFS) of color, texture (Gray-Level Co-occurrence Matrix), and shape-based features, achieving superior efficiency over complex Convolutional Neural Networks (CNNs). Unlike CNNs, which require 13 million trainable parameters, our methodology leverages as few as 10 IFS features to deliver 95% accuracy and an ROC-AUC score of 0.98. We created a diverse dataset of 2227 white bread images, classified into 1142 edible and 1085 inedible samples, to enhance model robustness. A tailored feature engineering pipeline, involving rigorous empirical evaluation and filtering of irrelevant features (e.g., constant, correlated), optimizes the IFS for binary classification. This approach outperforms state-of-the-art methods, offering a computationally efficient solution suitable for real-time bakery quality control. However, the dataset, primarily sourced from local bakeries, may introduce bias, limiting generalizability to other bread types. These results highlight the potential of machine learning for reliable, automated food safety assessments in resource-constrained settings.