Iranian Scientific Fisheries Journal

fa مقاله علمی – پژوهشی:‌ مدل‌سازی و پیش‌بینی عمر ماندگاری میگوی Litopenaeus vannamei با رویکردهای مبتنی بر یادگیری ماشین Modeling and predicting the shelf life of Litopenaeus vannamei with machine learning-based approaches بيوتكنولوژي و فرآوري هاي شيلاتي بيوتكنولوژي و فرآوري هاي شيلاتي پژوهشي Research فساد غذاهای دریایی، به دلیل رشد سریع میکروارگانیسم‌ها و به دنبال آن تغییر در طعم و بو، نگرانی‌های زیاد مصرف‌کنندگان را به‌دنبال دارد. هدف پژوهش حاضر، توسعه مدل‌های یادگیری ماشین برای پیش‌بینی شاخص‌های فساد میکروبی و شیمیایی میگوی Litopenaeus vannamei در شرایط مختلف نگهداری بود. مطالعه حاضر به بررسی توسعه مدل‌های مربوط به پیش‌بینی شاخص‌های میکروبی و شیمیایی با استفاده از الگوریتم‌های یادگیری ماشین (Random Forest، Extra Tree، k-Nearest Neighbors، AdaBoost، Gradient Tree Boosting و <a name="_Hlk217028963">Regression</a> Support Vector)، برای ارزیابی عمر ماندگاری L.vannamei (50 قطعه) پرداخت. چگونگی افزایش داده‌ها در یک مطالعه آزمایشگاهی (با 3 تکرار برای هر تجزیه‌‌وتحلیل) برای ارزیابی عمر ماندگاری میگو طی 15 روز، در دماهای مختلف (0، 4 و 8 درجه سانتی‌گراد)، چگونگی اثبات سازگاری داده‌های افزایش یافته با داده‌های اصلی و چگونگی بهینه‌سازی روش‌های موثر یادگیری ماشین برای فرآیندهای حل مسئله ارزیابی گردید. برای هر یک از شاخص‌های Pseudomonas، بازهای نیتروژن فرارکل (TVB-N) و میزان باکتری‌های هوازی (TVC)، تعداد 158 داده تولیدی وارد فرآیند مدل‌سازی شد. بنابراین، مجموعه داده نهایی شامل 474 داده تولیدی برای سه شاخص بود. میزان Pseudomonas، TVC و میزان  TVB-Nبا گذشت زمان نگهداری و دما افزایش یافت و بالاترین مقدار آنها به‌ترتیب log cfu/g 9/7 ،log cfu/g 7/8 و 73 میلی‌گرم بر 100 گرم گوشت میگو رسید. نتایج نشان داد که الگوریتم بهینه‌شده Regression Support Vector برای تخمین میزان Pseudomonas با مقدار R2 برابر با 9828/0، تخمین مقدار TVC با R2 برابر با 9928/0 و R2 برابر با 9940/0 برای TVB-N موثرترین مدل بود درحالی‌که سایر الگوریتم‌ها نسبت به این الگوریتم تاثیر کمتری داشتند. این نتایج نشان می‌دهد که توسعه مدل‌های یادگیری ماشین برای پیش‌بینی شاخص‌های فساد میکروبی و شیمیایی میگوی وانامی در شرایط مختلف نگهداری موثر بود. علاوه‌براین، باید تأکید کرد که تعداد فراسنجه‌‌های ورودی دما، تعداد باکتری‌ها، TVB-N برای افزایش داده‌ها برای توسعه و کاربرد الگوریتم‌های یادگیری ماشین بسیار مهم هستند. Introduction This paper focuses on the importance of shrimp quality, which is a favorite among consumers worldwide. However, seafood can quickly spoil under unfavorable storage conditions due to its nutrient content, water content, and weak connective tissues. Spoilage of seafood results from the rapid proliferation of microorganisms in seafood stored under inappropriate conditions, especially with the effect of temperature, causing changes contrary to consumer preferences, especially odor in seafood. It has been reported that spoilage microorganisms in seafood generally belong to Pseudomonas spp., Enterbacteriaceae, Vibrio sp, Lactobacillaceae, and Bacillaceae groups. The most important challenge in the seafood industry is their high spoilage potential (Shamaei and Roomiani, 2021). Predictive microbiology can determine the dynamics of microorganisms with respect to environmental factors and is one of the most widely used rapid methods to determine the shelf life, quality, and safety of seafood. Predicting the shelf life of seafood is crucial to maintaining their quality and safety throughout the supply chain (Kumarave et al., 2025). Recent advances in non-destructive technologies such as spectral imaging and nuclear magnetic resonance spectroscopy have overcome some of these limitations, but their widespread application remains a challenge. To overcome these obstacles, the seafood industry is increasingly exploring the use of machine learning and deep learning techniques, which have shown great potential in assessing food quality, reducing subjectivity, and enabling real-time decision-making (Rokhva et al., 2024). These technologies can improve the accuracy and efficiency of seafood freshness detection while minimizing manual intervention and providing much faster results compared to traditional methods (Cui et al., 2024). Machine learning algorithms that can predict the quality and shelf life of seafood products are also used in predictive microbiology. Machine learning algorithms have attracted much attention in various fields due to their advanced feature learning capabilities, making them a promising alternative for improving model performance in food quality and safety assessment. By reviewing studies, it was observed that algorithms such as Random Forest, Decision Tree, k-Nearest Neighbors, AdaBoost, and Gradient Tree Boosting can predict seafood quality (Gürfidan et al., 2024). Methodology Pseudomonas spp. and Total Viable Count (TVC) values were analyzed to determine. The amount of TVB-N in shrimp muscle was measured by distillation method according to the standard (AOAC, 1995). This study investigated predictive microbiology models using machine learning algorithms such as Random Forest, Extra Tree, k-Nearest Neighbors, AdaBoost, Gradient Tree Boosting, and Support Vector Regression to predict the shelf life of shrimp. The success metrics and test graphs obtained because of these trainings were interpreted and the most successful algorithm from the experiments was determined as the Support Vector Regression model. The inputs of the machine learning model are time and temperature values. It was evaluated how to augment data in a laboratory study to evaluate the shelf life of shrimp at different temperatures, how to prove the consistency of augmented data with the original data, and how to optimize successful machine learning methods for problem-solving processes. Based on these values, TVC, Pseudomonas spp. and TVB-N values are estimated. The results obtained provide powerful data for the prediction of the shelf life of shrimp. Results According to the results of the research, the initial number of the bacteria varied between 2.7 and 3.5 log cfu/g in whole shrimp stored at different temperatures. The highest initial value was observed at 10°C. Pseudomonas spp. numbers increased with storage time and the highest value was reported as 7.9 log cfu/g. Ntzimani et al. (2023) investigated the effect of ice on transportation and storage quality in seabass fish. The researchers reported an initial Pseudomonas spp. value of 2.0 log cfu/g in shrimp, lower than the values obtained in this study where different algorithms were tested. Figure 1 shows the scatter plot of the original data set, where a part of the data set is Pseudomonas spp., b part of the data set is TVC bacteria and c part of the data set is TVB-N. Figure 2 shows the scatter plot of the new data set after the data augmentation was applied. When analyzed figuratively, it is seen that the trend line on the temperature, time, and log graphs is quite close to each other in both. The small differences observed in the column graphs will be explained by the numerical values measured in the similarity metrics. Data augmentation has been reported by researchers to be a method that plays a role in predicting data at points that are not tested and in making more accurate and reliable classification. Discussion and conclusion These results indicate that machine learning methods can be a fast, powerful and effective tool for predicting shrimp shelf life. In addition, it should be emphasized that the number of input parameters such as temperature, number of bacteria, volatile nitrogenous bases are very important to increase the data for the development and application of machine learning algorithms. Gowda et al. (2023) investigated the determination of freshness in edible seafood using IoT and machine learning techniques. The researchers tested machine learning algorithms such as Dense Net Algorithm and Efficient Net Algorithm while determining the freshness of the seafood products they used. According to the results of the research, Efficient Net and Dense Net Algorithm reported accuracy rates of 0.085 and 0.075, respectively. Compared to the studies in the literature, it was observed that machine learning algorithms produced successful results in determining the quality, shelf-life prediction, and freshness classification of seafood products, although there were methodological differences in this study. Recently, Hou et al. (2025) performed non-destructive detection of quality changes in shrimp using the YOLO algorithm. The researchers obtained high correlations of 0.88 and 0.81 through the TVB-N and TVC levels of shrimp. In another study, researchers aimed to determine the freshness values of various seafood products in real time. In this context, the authors used four different machine learning algorithms, namely neural network, genetic algorithm, extreme learning machine, and radial basis function neural network. According to the results of the study, it was reported that the genetic algorithm and neural network models showed higher accuracy in predicting the shelf life of various seafood products. Among the four different machine learning algorithms, it was also reported that the neural network model had an R2 value of 0.99 and was shown to be the best model fit (Cui et al., 2024). Conflict of interest The authors declare no conflict of interest. Acknowledgment We should acknowledge the necessity of thanking the staff of the Fisheries lab of the Islamic Azad University for their assistance in analysis. Litopenaeus vannamei, ماندگاری, یادگیری ماشین, Regression Support Vector Litopenaeus vannamei, Shelf life, Machine Learning, Support Vector Regression 43 73 http://isfj.ir/browse.php?a_code=A-10-206-15&slc_lang=fa&sid=1 Asma Kalantari اسماء کلانتری asmakalantri45@gmail.com 100319475328460042731 100319475328460042731 No Department of Food Science and Technology, Ahv. C., Islamic Azad University, Ahvaz, Iran گروه علوم و صنایع غذایی، واحد اهواز، دانشگاه آزاد اسلامی، اهواز، ایران Laleh Roomiani لاله رومیانی Laleh.roomiani@iau.ac.ir 100319475328460042732 100319475328460042732 Yes Department of Fisheries, Ahv. C., Islamic Azad University, Ahvaz, Iran گروه شیلات، واحد اهواز، دانشگاه آزاد اسلامی، اهواز، ایران Mehrnoosh Tadayoni مهرنوش تدینی me.tadayoni@iau.ac.ir 100319475328460042733 100319475328460042733 No Department of Food Science and Technology, Ahv. C., Islamic Azad University, Ahvaz, Iran گروه علوم و صنایع غذایی، واحد اهواز، دانشگاه آزاد اسلامی، اهواز، ایران