Purpose: This study aimed to develop a web-based system to detect the freshness level of Cavendish bananas using image-based classification to help assess fruit quality and reduce losses during distribution.

Methodology/approach: This study used a dataset of 1,030 labeled images of Cavendish bananas categorized into four classes: fresh, ripe, unripe, and rotten. Image preprocessing techniques, such as resizing, normalization, and data augmentation, were applied. A Convolutional Neural Network (CNN) model was trained using Python with machine learning libraries, and the trained model was integrated into a web application built with Streamlit to enable real-time image-based predictions.

Results: The proposed CNN model achieved an accuracy of 99%, with average precision, recall, and F1-score values of 0.99. The web-based system successfully provided real-time freshness predictions through image uploads or camera captures.

Conclusions: This study confirms that the web-based CNN system effectively detects Cavendish banana freshness with high accuracy. The integration into a Streamlit application enables real-time and practical use for quality assessment. The system demonstrates the potential of machine learning to support agricultural quality control, reduce post-harvest losses, and enhance food safety.

Limitations: This study was limited to Cavendish banana images and controlled image conditions; therefore, performance may vary when applied to other banana varieties or different lighting environments.

Contributions: This study contributes to the application of machine learning and computer vision in agricultural quality assessment. The system provides a practical tool for farmers, distributors, and consumers to objectively evaluate banana freshness, reduce economic losses, and support public health.