<![CDATA[Black mold growing on damp indoor surfaces produces harmful mycotoxins that pose serious health risks, yet conventional laboratory-based identification remains slow, costly, and largely inaccessible for rural communities. This study aims to develop Peka Toksin, an Android-based mobile system that integrates a Convolutional Neural Network (CNN) with a Management Information System (MIS) to support real-time, community-level early detection of black mold. The classifier was built using transfer learning on the MobileNetV2 architecture, trained on a combined dataset of 10,537 labeled images (7,642 black-mold and 2,895 non-black-mold) obtained through field documentation during a community service program in Pamekaran Village, West Bandung Regency, supplemented with publicly available mold-image collections, and split into 7,569 training and 2,968 validation images. The application itself was developed using the Waterfall model, covering analysis, design, implementation, testing, and maintenance. Evaluation on the validation set showed an overall accuracy of 88%, with precision of 0.81, recall of 0.99, and F1-score of 0.89 for the Black Mold class, and precision of 0.98, recall of 0.76, and F1-score of 0.86 for the Non-Black Mold class, indicating high sensitivity to black mold with a tendency toward false positives. Beyond classification, Peka Toksin provides a detection-history feature and an action guide to support ongoing monitoring and decision-making. These results indicate that Peka Toksin can serve as an accessible, low-cost early-warning tool for black mold exposure, while its MIS component enables structured data management to support future village-level public-health surveillance.]]>
Copyrights © 2025
