Classification of Insidious Fruit Rot (IFR) Stages in Harumanis Mango by Utilising Vibration-Based Sensors with Machine Learning

Abstract

Insidious Fruit Rot (IFR) significantly impacts the quality and marketability of Harumanis mangoes (Mangifera indica L.), with traditional manual inspection methods being labor-intensive and error-prone. To address these limitations, this study proposes an automated detection system integrating vibration-based sensors with machine-learning models for precise IFR stage classification. Data collection involved piezoelectric vibration sensors and electret microphones, followed by pre-processing and feature extraction. Principal Component Analysis (PCA) was employed to reduce data dimensionality while preserving key information. Machine learning models, including Random Forest (RF) and Gradient Boosting (GB), were trained and evaluated using precision, recall, F1 scores, and accuracy metrics. A Voting Classifier, combining outputs from RF and GB models, achieved an overall accuracy of 85%. Performance metrics for IFR stages were as follows: Non IFR (Precision: 0.75, Recall: 1.00, F1-score: 0.86), Minor IFR (Precision: 1.00, Recall: 0.88, F1-score: 0.93), Major IFR (Precision: 1.00, Recall: 0.33, F1-score: 0.50). Example classifications demonstrated effective differentiation between IFR stages. This study highlights the potential of integrating sensor technology with machine learning for real-time IFR detection, enabling improved quality control and efficiency in agriculture. Future research will optimize models, incorporate additional sensors, and validate the system in real-world applications.