Optimizing Gas Sensor Array in an Electronic Nose Using Firefly Algorithm with VGG-1D for Cookie Cooking Level Classification

During electric oven baking, cookie cooking levels are assessed visually by color; however, this can be inaccurate when colors appear similar. Aroma provides a more reliable marker through volatile organic compounds, measurable with electronic nose systems. While using many gas sensors increases cost and complexity, reducing their number requires careful selection and robust machine learning to maintain performance. To address these limitations, the novelty of this study lies in applying the firefly algorithm (FA) for optimal gas sensor selection, guided by brightness-based attractiveness and movement to reduce the number of sensors while preserving accuracy, and in adapting modern machine learning models-visual geometry group (VGG), residual network (ResNet), Inception, and EfficientNet-into one-dimensional (1D) for cookie cooking level classification during baking. Results show that FA selected five sensors, with VGG-1D achieving an F1-score of 96.25%, demonstrating that FA-based sensor selection with modern machine learning enhances detection and supports quality consistency.