Abdul Kareem, Razia Sulthana ORCID: 0000-0001-5331-1310 , Frost, Nageena, Goodall, Iain, Tilford, Timothy ORCID: 0000-0001-8307-6403 and Palacios, Ana Paula (2024) Machine learning to detect fungal infections in stored pome fruits via mass spectrometry data: industry, economic, and social implications. Journal of Advances in Information Technology, 15 (10). pp. 1174-1183. ISSN 1798-2340 (Print), 1798-2340 (Online) (doi:https://doi.org/10.12720/jait.15.10.1174-1183)

Preview

PDF (Open Access article) 48410 ABDUL KAREEM_Machine_Learning_To_Detect_Fungal_Infections_In_Stored_Pome_Fruits_Via_Mass_Spectrometry_Data_Industry_Economic_And_Social_Implications_(OA)_2024.pdf - Published Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (2MB) | Preview

Abstract

Pome fruits, notably apples and pears, experience decay during storage due to fungal infections. The timely discernment of these infections is imperative to avert the deterioration of these fruits within warehouse confines. In an experimental setup, two distinct apple cultivars, Braeburn and Gala, were inoculated with fungi Monilinia laxa, Neonectria ditissima, and Botrytis cinerea. As the infection progresses, the apples release chemical volatile components, which are measured using mass spectrometry in both positive and negative ion modes, recording mass-charge ratios ranging from m/z 30 to m/z 900 with a 0.3 Dalton difference between each measurement. The dataset is then partitioned into 24 sets of three-dimensional data, encompassing attributes related to two types of apples, three types of fungi, and two types of ions. They are analyzed using various machine learning algorithms, including Logistic Regression, Support Vector Machines, XGBoost, Random Forest, and four distinct customised Neural Networks, to classify infected and uninfected apples. The outcomes from the different machine learning algorithms across the 12 combinations of Apple-FungiIon are recorded, revealing that certain algorithms excel in different combinations. The performance metrics namely True Positive, True Negative, False Positive, False Negative, Accuracy are closely analysed and the algorithms that produces the highest and second-highest accuracy are highlighted. Upon thorough analysis of the 12 combinations, it is observed that Logistic Regression and SVM with a linear kernel achieve the highest accuracy in approximately 11 combinations. Specifically, Logistic Regression achieves a precision of 98% for Braeburn apples, while SVM attains a 99% accuracy for Gala apples. This research project has a triple impact on industry, economy, and society. On an industrial level, the precision and early predictions of the proposed work can effectively safeguard large quantities of apples in storage bins. Economically, it has the potential to avert substantial monetary losses. Societally, it plays a crucial role in determining the ideal timing to release fruits to the market for consumption without jeopardizing human health.

Item Type:	Article
Uncontrolled Keywords:	pome fruits, apples, fungal infections, mass spectrometry, machine learning, neural networks
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA75 Electronic computers. Computer science S Agriculture > S Agriculture (General)
Faculty / School / Research Centre / Research Group:	Faculty of Engineering & Science Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified:	23 Oct 2024 15:43
URI:	http://gala.gre.ac.uk/id/eprint/48410

Actions (login required)

View Item

Downloads