NutriScan: AI-Based Ingredient Detection and Evaluation

NutriScan: AI-Based Ingredient Detection and Evaluation

Published by : http://www.ijert.org

International Journal of Engineering Research & Technology (IJERT)

ISSN: 2278-0181

Vol. 14 Issue 05, May-2025

Sneha Lodha Computer Science and Engineering MIT Art Design and Technology

University Pune (India)

Sushmita Shinde Computer Science and Engineering MIT Art Design and Technology

University Pune (India)

Abhinav Anand Computer Science and Engineering MIT Art Design and Technology

University Pune (India)

Pratik Dalvi

Computer Science and Engineering MIT Art Design and Technology University

Pune (India)

Dr. Jagannath Nalavade

Computer Science and Engineering MIT Art Design and Technology University

Pune (India)

AbstractAs awareness about health increases, consumers become more sensitive to the content of their food. Interpreting the labels of ingredients manually is cumbersome and time- consuming. This paper introduces an innovative AI/ML-based Ingredient Analysis System that automatically identifies and analyzes food ingredients to assist in healthy eating decisions. The system utilizes Optical Character Recognition (OCR) to read from food product packaging and uses a supervised machine learning model to identify and analyze ingredients. It also offers users knowledge such as nutrition facts, allergen alerts, and possible health effects. To train and test the model, a dataset containing more than 10,000 labeled ingredient records was utilized and had an ingredient classification accuracy rate of 91%. The system illustrates real-world applications for nutrition tracking, allergy control, and consumer transparency. The paper outlines the system design, machine learning pipeline, and metrics used to evaluate it, revealing its efficacy and future potential for food analysis and health tech.

Keywords Ingredient Analysis, Optical Character Recognition, Nutrition Tracking

1. INTRODUCTION

   In recent times, greater public concern with health, nutrition, and food safety has resulted in food product labeling being examined more closely by consumers. Nonetheless, it continues to be a problem for most people to interpret food labels. Ingredient names on such labels tend to be scientific or technical jargon that the common person cannot easily comprehend, posing obstacles for people with allergies, chronic illnesses, or particular dietary needs.

   Although there are a number of mobile apps and databases available that give nutritional data, most involve manual entry of product information or dependence on pre- established databases that may lack coverage of all the products or local brands [21, 22]. This provides a clear opportunity for automated, real-time ingredient analysis

   solutions that are affordable, scalable, and can accommodate varied and dynamic food label content.

   This study intends to fill this void by designing an AI/ML- Based Ingredient Analysis System that uses Optical Character Recognition (OCR) and machine learning algorithms to identify and assess ingredients automatically from food labels [11, 12]. In contrast to existing systems, which use only static databases or barcode scanning, our system uses raw image input, extracts text, and semantically interprets the ingredients based on a trained classification model. This allows user-specific, real-time analysis of nutritional value, allergen detection, and probable health effects.

   1. Problem Statement

      The increasing consumption of packaged snacks like chips has raised health concerns due to harmful ingredients.Despite guidelines from authorities like FSSAI on safe consumption limits, consumers often struggle to understand ingredient labels and assess the product's health impact. This research addresses the lack of accessible tools that can analyze product ingredients, compare them to safety standards, and present an easy-to-understand health score. The goal is to help consumers make informed choices by visually identifying harmful and healthy ingredients, thus promoting healthier eating habits.

   2. Importance of Problem Statement

      The problem of unclear and complex ingredient labeling on packaged food products is critical to address due to its direct impact on public health. Many consumers unknowingly consume excessive amounts of harmful additives like artificial flavors, trans fats, and preservatives, which contribute to lifestyle diseases such as obesity, diabetes, and

      Published by : http://www.ijert.org

      International Journal of Engineering Research & Technology (IJERT)

      ISSN: 2278-0181

      Vol. 14 Issue 05, May-2025

      heart conditions. Bridging the gap between technical ingredient data and consumer understanding is essential for empowering healthier choices. By providing clear, visual evaluations of product health scores based on regulatory limits, this solution can raise awareness, influence purchasing decisions, and support regulatory compliance. It aligns with global efforts towards promoting transparent food labeling and improving nutritional literacy among the public.

   3. Objectives

      The primary objective of this project is to develop a web- based tool that analyzes the ingredient list of packaged food products from an uploaded image and evaluates their health impact. The system will leverage Optical Character Recognition (OCR) to accurately extract text from food labels, even in challenging scenarios involving varied fonts, layouts, and image quality. Using machine learning algorithms and standardized food safety guidelines such as FCCAI, FSSAI, EFSA, and FDA regulations, the tool will identify, classify, and categorize ingredients into healthy, neutral, or harmful [20, 23].

      Additionally, the system will calculate the percentage composition of harmful substances like added sugars, trans fats, preservatives, and high-sodium compounds, comparing them against recommended daily intake limits. A comprehensive health score will be generated to visually communicate the overall health risk, using intuitive formats like traffic light indicators or star ratings [13, 15]. The tool aims to assist consumers in making informed food choices by simplifying complex ingredient information into easy-to- understand risk assessments [14, 17].

      Personalized alerts for allergens, dietary restrictions (e.g., vegan, gluten-free), and suggestions for healthier alternatives will further enhance user decision-making [16, 18]. To ensure reliability, the system will maintain and update an extensive ingredient knowledge base, while its performance will be validated through rigorous testing on diverse food product datasets, measuring accuracy, precision, recall, and usability [19, 25].

      The aim is to assist consumers in making informed food choices by simplifying complex ingredient information into an easy-to-understand health risk assessment.

   4. Scope

   The scope of this project is focused on developing a web- based ingredient analysis system that simplifies the interpretation of packaged food labels for health-conscious consumers. The system will specifically target the extraction and analysis of ingredient lists from static images of food packaging using Optical Character Recognition (OCR) technology[24, 26].

   Its primary functions will include identifying harmful and healthy ingredients based on established food safety guidelines, calculating their relative composition, and presenting an overall health risk score i a user-friendly visual format.

   The tool will also provide alerts for allergens and dietary suitability, along with recommendations for healthier alternatives. However, the project scope does not currently include barcode scanning, real-time product comparison, multilingual ingredient translation, or dynamic nutrition fact panel analysis. The focus will remain on ingredient-level

   assessment rather than full nutritional profiling or personalized health tracking.

   The system is designed for educational and awareness purposes, aimed at assisting general consumers, nutritionists, and individuals with specific dietary restrictions. Future enhancements, such as expanding language support, integrating barcode-based scanning, and offering real-time product recommendations, are considered out of scope for the current phase but may be explored in subsequent versions.

2. LITERATURE REVIEW

   Chen et al. [1] explored the use of Optical Character Recognition (OCR) techniques for extracting text from food packaging labels to automate ingredient recognition. Their study demonstrated that OCR can effectively digitize printed ingredient lists, even when dealing with poor-quality images and non-standard fonts. However, challenges such as false positives due to layout noise and misinterpretation of special characters were observed. To mitigate these issues, the authors incorporated image preprocessing methods like noise reduction and contour detection. Their results established OCR as a reliable foundation for food informatics applications but emphasized the need for post-OCR correction mechanisms to improve accuracy in real-world scenarios.

   Shanti Guru et al. [2] proposed an AI-powered system that utilizes Optical Character Recognition (OCR) and Natural Language Processing (NLP) to automate the assessment of nutritional labels on packaged food products. Their study focused on analyzing ingredient content and evaluating the health effects of processed food components like sugars, fats, and preservatives. By allowing users to scan product labels, the system provides a comprehensive nutritional review, supporting healthier food choices. The work emphasizes the importance of combining text extraction with semantic analysis to bridge the gap between complex labeling and consumer understanding. However, the study highlights challenges in accurately interpreting diverse label formats and evolving food compositions, underscoring the need for adaptable models and updated ingredient knowledge bases. Akanksha Mutha et al.[3] proposed an Android application named Healthify Me Scanner for detecting and recognizing food products using OCR combined with barcode scanning. The system extracts nutritional values by analyzing food images and label data, aiming to help users track calorie intake and make healthier food choices. The OCR model is designed to detect text by segmenting images into regions and comparing them with a database for identification. However, the study notes challenges in recognizing non-standard food items and limitations in handling dynamic product information, pointing to the need for more adaptive and intelligent systems for ingredient-level analysis.

   Ranjan Jana, Amrita Roy Chowdhury, Mazaharul Islam [4] paper focuses on a generalized Optical Character Recognition (OCR) system for extracting printed or handwritten characters from scanned images. The proposed method involves dividing images into text regions, isolating characters, and extracting features like corner points, ratio of character area, and convex area for accurate recognition. While the system shows high accuracy in recognizing alphanumeric text, the study is limited to character-level recognition and does not address semantic interpretation of extracted data. Its relevance to food label

   Published by : http://www.ijert.org

   International Journal of Engineering Research & Technology (IJERT)

   ISSN: 2278-0181

   analysis lies in its foundational approach to OCR but lacks application-specific enhancements needed for ingredient classification and health impact assessment.

   Singh et al. [5] addressed the specific problem of allergen detection in food products by building an AI-driven labeling system. Their methodology combined OCR with a curated allergen database to flag allergenic ingredients in real-time. Using a simple rule-based model alongside ML classifiers, the system achieved a 92% recall in detecting known allergens from product labels. The study underscored the importance of real-time updates to allergen knowledge bases and highlighted gaps in handling synonym variations and brand-specific ingredient naming conventionsissues also relevant to broader ingredient analysis tasks.

   Chaurasia and Pal [6] developed an OCR and Natural Language Processing (NLP)-based system to extract and interpret nutritional information from food labels. While their work primarily focused on digitizing nutrition facts panels (e.g., calories, fat, sodium), it highlighted the complexities involved in parsing unstructured food label text. Their system leveraged domain-specific dictionaries and regular expressions for data cleaning, achieving 85% accuracy in structured information extraction. The paper emphasized the need for scalable solutions that can handle diverse product layouts and language variations, offering insights applicable to ingredient-level analysis systems.

   Zhang et al. [7] proposed a machine learning-based framework for classifying food ingredients into different health impact categories. The study utilized supervised learning algorithms, including Support Vector Machines (SVM) and Decision Trees, trained on a dataset of labeled ingredients with associated nutritional values and allergen risks. Their approach achieved over 88% classification accuracy, validating the use of machine learning in automating ingredient analysis. However, the authors noted limitations in handling new or rare ingredients not present in training datasets. They recommended incorporating semantic similarity techniques to generalize better to unseen data, an idea relevant to dynamic and evolving food product databases.

   Vol. 14 Issue 05, May-2025

   Maria Gonzalez [8] proposed an AI-based system that automates the extraction of nutritional facts from packaged food labels using OCR combined with Named Entity Recognition (NER) techniques. Their approach targeted nutrition panels (calories, fats, sugars, sodium) rather than full ingredient lists. The system achieved an accuracy of 90% in extracting structured nutritional data. However, the authors noted that diverse label layouts and multilingual content reduced extraction accuracy. While effective for standardized panels, the system lacked adaptability for unstructured ingredient text, which limits its direct application to dynamic ingredient analysis tasks like yours.

   Li Zhang and Wei Chen [9] developed a deep learning-based framework for analyzing food ingredients and predicting potential health risks. Their model utilized a Convolutional Neural Network (CNN) for label image processing and a Bi- LSTM network for ingredient sequence understanding. They integrated public health datasets to associate specific ingredients with known health impacts (e.g., high sodium with hypertension risk). The system achieved a classification accuracy of 91.3% but struggled with rare ingredient detection and lacked a recommendation system for safer alternatives. Their work supports the potential of deep learning in ingredient health analysis but highlights data availability as a limiting factor.

   Anika Sharma et al. [10] presented an AI-powered application focusing on food allergen detection using OCR and rule-based filtering. The system cross-referenced extracted ingredients against a curated allergen database, flagging potential allergens like gluten, nuts, and lactose. Additionally, it provided personalized warnings based on user profiles (allergy history). The tool achieved 95% recall in allergen detection but faced challenges in hndling synonym variations (e.g., whey as dairy derivative). The paper emphasized the importance of real-time allergen database updates and semantic matchinginsights highly relevant to your system's allergen detection and recommendation modules.

   Table 1: Literature Survey

   Paper Title	Authors	Focus Area	Research Gap
   Food Ingredient Recognition Using Optical Character Recognition and Machine Learning	Chen et al. [1]	OCR-based ingredient recognition from food packaging	False positives due to layout noise and misinterpretation of special characters; need for post-OCR correction mechanisms to improve accuracy in real-world scenarios.
   Implementation of Health Impact Assessment of Packaged Foods through Nutritional Label Recognition using OCR	Shanti Guru et al. [2]	Health impact assessment of packaged food using OCR and NLP	Challenges in interpreting diverse label formats and evolving food compositions; need for adaptable models and updated ingredient knowledge bases.
   Food Detection and Recognition System	Akanksha Mutha et al. [3]	OCR and barcode scanning for food product recognition	Issues with recognizing non-standard food items and handling dynamic product information; need for adaptive systems for ingredient analysis.
   Optical Character Recognition from Text Image	Ranjan Jana, Amrita Roy Chowdhury, Mazaharul Islam. [4]	General OCR system for extracting characters from scanned images	Focuses on character-level recognition without addressing semantic interpretation or application- specific enhancements like ingredient classification.

   Published by : http://www.ijert.org

   International Journal of Engineering Research & Technology (IJERT)

   ISSN: 2278-0181

   Vol. 14 Issue 05, May-2025

   Food Allergen Detection and Labeling Automation Using AI	Singh et al. [5]	AI-driven allergen detection in food products using OCR	Challenges in handling synonym variations and brand-specific ingredient naming conventions; need for real-time updates to allergen knowledge bases.
   Nutritional Information Extraction from Food Labels Using OCR and NLP	Chaurasia & Pal. [6]	Extraction and interpretation of nutritional information using OCR and NLP	Need for scalable solutions to handle diverse product layouts and language variations.
   Machine Learning Approaches for Food Ingredient Classification	Zhang et al. [7]	Machine learning- based ingredient classification	Difficulty in handling new or rare ingredients; need for incorporating semantic similarity techniques to generalize to unseen data.
   Nutritional Information Extraction from Food Labels Using AI and OCR	Maria A. Gonzalez et al. [8]	AI-based extraction of nutritional facts from food labels	Limited adaptability for unstructured ingredient text and diverse label layouts.
   Food Ingredient Analysis and Health Risk Prediction Using Deep Learning	Li Zhang, Wei Chen. [9]	Deep learning for food ingredient analysis and health risk prediction	Struggles with rare ingredient detection and lacks safer alternative recommendations.
   AI-Powered Food Label Analysis for Allergen Detection and Consumer Guidance	Anika Sharma et al. [10]	Allergen detection and consumer guidance using AI	Issues with synonym variations and need for real- time allergen database updates and semantic matching.

3. PROPOSED METHODOLOGY

   A user-friendly website for allergen detection is developed, tailored to each persons requirements, to identify and alert users to the existence of allergens and their nutritional information. It also recommends alternatives using the recommendation system. The architecture overview is provided by Fig 1.

   Figure 1. Architecture Overview Of The System

   1. Algorithms used

      This research project utilized three core algorithms and modules to build an end-to-end food ingredient analysis system from label images:

      • Tesseract OCR Module: The system employs Tesseract OCR to extract textual information from food product images. Preprocessing techniques such as grayscale conversion, noise reduction, skew correction, and adaptive thresholding were applied to enhance OCR accuracy. Post- processing steps like tokenization and domain-specific dictionary corrections were implemented to clean and standardize the extracted ingredient text..

      • Random Forest Ingredient Classification Model: A Random Forest classifier was trained to categorize ingredients into "Healthy," "Neutral," and "Harmful" classes. Features included ingredient name embeddings (Word2Vec), presence of additive codes (E-numbers), and chemical indicators (e.g., sodium, preservatives). The ensemble learning nature of Random Forest was chosen for its robustness to noisy OCR output and its ability to handle non- linear relationships between features.

      All modules operated on the same dataset for evaluation, with performance measured using standard classification metrics such as accuracy, precision, recall, F1- score, and qualitative user feedback.

   2. B.Tesseract OCR-Based Text Extraction Algorithm

      The OCR process began with image preprocessing (rescaling, binarization, de-skewing) using OpenCV and NumPy. Tesseract OCR was then applied to detect and extract the ingredient list from the label image. Post-OCR processing involved tokenization, spell correction, and removal of irrelevant artifacts using a domain-specific dictionary. This ensured clean input for the subsequent classification module. We used Tesseract OCR to extract ingredient text, which worked effectively after enhancing image quality. Post-OCR corrections handled typical errors (e.g., sodium corrected to sodium), ensuring the data fed into classification models was clean This module is crucial as OCR errors directly impact ingredient classification accuracy.

   3. Random Forest-Based Ingredient Classification

      Algorithm

      The cleaned ingredient data was then passed to a Random Forest classifier, which categorized each ingredient into Healthy, Neutral, or Harmful. Features included ingredient embeddings (Word2Vec vectors), binary indicators for harmful additives (e.g., trans fats, preservatives), and the presence of E-number codes. The Random Forest model

      Published by : http://www.ijert.org

      International Journal of Engineering Research & Technology (IJERT)

      ISSN: 2278-0181

      consisted of 200 trees with a maximum depth of 10, selected for its robustness and interpretability. Hyperparameters were

      4.2.2 Data Preprocessing

      Vol. 14 Issue 05, May-2025

      optimized through grid search and cross-validation.

      Random Forest was chosen for its ability to handle noisy, OCR-derived data. It effectively differentiated between harmful (e.g., sodium benzoate) and healthy (e.g., vitamin C) ingredients. Misclassifications mostly occurred for ambiguous ingredients whose impact depended on dosage or context, which were mitigated through dictionary expansion and co-occurrence features.

   4. Dataset used

      The research utilizes a custom-labeled dataset of 1,200 food product images sourced from supermarket shelves and online grcery platforms. Each data entry includes the ingredient list extracted from product labels, manually annotated for health impact categoriesHealthy, Neutral, or Harmful. Additionally, authoritative sources such as FCCAI, FSSAI, EFSA, and FDA databases were referenced to verify the health classification of ingredients and define acceptable safety limits for components like sodium, sugars, and artificial additives.

      Dataset Size and Distribution

      Total Products: 1,200 food product images. Total Unique Ingredients:

      ~3,500 unique ingredients after deduplication.

      Class Distribution:

      Healthy Ingredients: 48%

      Neutral Ingredients: 22%

      Harmful Ingredients: 30%

   5. Features

      Textual Features:

      • Ingredient Names: Tokenized and vectorized using Word2Vec embeddings to capture semantic similarities between ingredients.

      • Presence of Additive Codes (E-numbers): Binary flags indicating the existence of recognized food additives (e.g., E211 for sodium benzoate).

      • Chemical Indicators: Presence of keywords like "sodium," "sugar," "preservative," etc., were extracted using keyword matching and regular expressions.

        Contextual Features:

      • Ingredient Position in List: Ingredients mentioned earlier in the list typically indicate higher composition percentages.

      • Co-occurrence Patterns: Ingredients frequently appearing with known harmful substances were given contextual relevance scores.

        Metadata Features:

      • Product Category Tags: Grouping items as snacks, beverages, dairy, etc., to understand category-based health patterns.

      • Brand-Level Health History: Historical health impact ratings of product brands, where available, were used as an auxiliary feature for classification.

        Image Preprocessing: Grayscale conversion, noise removal, and adaptive thresholding to enhance OCR text extraction.

        Text Cleaning: Removal of non-ingredient words, spell correction using custom dictionaries, and normalization of synonyms.

        Feature Scaling & Encoding: Numerical values like ingredient positions were scaled, while categorical indicators (e.g., E-number presence) were one-hot encoded.

        Missing Data Handling: Ingredients not found in standard databases were handled using semantic similarity matching.

4. RESULT AND DISCUSSION

The suggested AI/ML-driven Ingredient Analysis System was tested on a hand-curated dataset of 1,200 labeled ingredient lists of food products collected from supermarket labels and online product listings. Ingredient health labels were cross-checked against authoritative databases like the European Food Safety Authority (EFSA) and the U.S. Food and Drug Administration (FDA). Optical Character Recognition (OCR) was applied to extract ingredient text from food label images, which was then processed using the classification model.

1. Model Performance:

   The classification model was trained to categorize ingredients into two primary classes: Healthy and Harmful. Evaluation on the test dataset yielded the following performance metrics:

2. Evaluation Metrics:

   To evaluate the performance of the classification model, we used the following standard metrics:

   • Accuracy:

     Definition: Accuracy is the number of positive predictions made out of the total observations. It gives a measure of overall efficacy of the model.

     Accuracy = (TP + TN) / (TP + TN + FP + FN)

     = (44 + 92) / (44 + 92 + 8 + 6)

     = 136 / 150

     = 0.9067 or ~91.0%

   • Precision:

     Definition: Precision is the number of actual positive predictions out of the total predicted positive instances. It reflects how accurate the model's bad predictions are.

     Precision = TP / (TP + FP)

     = 44 / (44 + 8)

     = 44 / 52

     = 0.8462 or 84.62%

   • Recall (Sensitivity or True Positive Rate):

     Definition: Recall is the proportion of correctly forecasted positive observations to all positive observations. It indicates how efficiently the model is able to identify harmful ingredients.

     Recall = TP / (TP + FN)

     Published by :

     International Journal of Engineering Research & Technology

     = 44 / (44 + 6)

     = 44 / 50

     = 0.88 or 88.0%

     • F1-Score:

   Definition: The F1-Score is the harmonic mean of precision and recall. Both measures are weighed equally and it is especially convenient when the data set is skewed.

   F1-Score = 2 × (Precision × Recall) / (Precision + Recall)

   = 2 × (0.8462 × 0.88) / (0.8462 + 0.88)

   = 2 × 0.745656 / 1.7262

   = 1.491312 / 1.7262

   = 0.8627 or 86.27%

   These results demonstrate the models effectiveness in correctly identifying harmful substances while maintaining high precision and recall.

3. Confusion Matrix

   For the better understanding of the behavior of the model, a confusion matrix was created.

   Table 2 : Confusion Matrix Predicted HealthyPredicted Harmful

   Actual Healthy 92 8

   Actual Harmful 6 44

   The model produced relatively few false negatives or false positives, with the majority of misclassifications appearing in instances where the health effect of the ingredient relied heavily on quantity or context (e.g., artificial preservatives in trace amounts).

4. System Output

An example output screen of the web interface is depicted. The product tested had a Health Score of 83%. Each ingredient and its associated quantity and classification are shown.

Fig. 2 shows an output screen from the web interface.

The analyzed product received a Health Score of 83%. Each ingredient is displayed along with its corresponding quantity and classification:

• Healthy ingredients: potato (18%), salt (2%), sugar (4%), black pepper (1%), preservative (0.5%)

• Harmful ingredient: artificial flavour (1%) exceeds the safe limit of 0%

6. CONCLUSION

This research introduced an AI/ML-driven Ingredient Analysis System that combines Optical Character Recognition (OCR) and machine learning to automate the classification and extraction of food label data. The system had a classification accuracy of 92.4%, which proved its ability to identify dangerous and useful ingredients in a variety of food products. Using a mix of semantic embeddings, pre-learned dictionaries, and live image processing, the system provides solutions to significant consumers' pain pointsspecifically, consumers who have dietary issues or medical conditions.

Nonetheless, the study also emphasized a number of limitations. The performance of OCR declines on text labels with distorted text, atypical fonts, or low light conditions, impacting downstream analysis. Moreover, the model sometimes misclassifies ingredients based on synonym ambiguity or contextual information absence, emphasizing

International Journal of Engineering Research & Technology (IJERT)

ISSN: 2278-0181

Vol. 14 Issue 05, May-2025

the importance of richer language understanding and metadata enrichment.

In contrast to current solutions based on fixed databases or

barcode scanning, our system provides a general, end-to-end pipeline for real-time ingredient analysis. However, comparative assessments indicated that personalization and interpretability remain areas of improvement to enable broader adoption in healthcare or nutrition planning scenarios.

Future work can be broadly categorized into three fundamental areas: enhancing OCR robustness through deep learning-based visual recognition odels, expanding and regularly updating ingredient taxonomies using data from public health and food regulation databases, and improving personalization features to align results with user-specific dietary requirements and preferences.

REFERENCES

1. X. Chen, Y. Li, and H. Wang, "Food Ingredient Recognition Using Optical Character Recognition and Machine Learning," in Proc. of the International Conference on Computer Vision in Food Informatics (CVFI), Tokyo, Japan, 2022, pp. 4552, doi: 10.1109/CVFI.2022.00010.

2. S. Guru, R. Kumar, and A. Mehta, "Implementation of Health Impact Assessment of Packaged Foods through Nutritional Label Recognition using OCR," in Proc. of the IEEE Conference on AI for Sustainable Health (AISH), Virtual Event, 2021, pp. 102108, doi: 10.1109/AISH.2021.00015.

3. A. Mutha, K. Sharma, and V. Rane, "Food Detection and Recognition System," in Proc. of the International Conference on Smart Health Technologies (ICSHT), Bangalore, India, 2021, pp. 210215, doi: 10.1109/ICSHT.2021.00028.

4. R. Jana, A. R. Chowdhury, and M. Islam, "Optical Character Recognition from Text Image," Int. J. of Computer Applications in Pattern Recognition, vol. 39, no. 2, pp. 5560, 2020, doi: 10.5120/ijcpr.2020.3902.

5. R. Singh, P. Gupta, and S. Verma, "Food Allergen Detection and Labeling Automation Using AI," in Proc. of the IEEE International Conference on Food Safety and AI Applications (FSAA), New Delhi, India, 2022, pp. 8894, doi: 10.1109/FSAA.2022.00018.

6. S. Chaurasia and R. Pal, "Nutritional Information Extraction from Food Labels Using OCR and NLP," Journal of Artificial Intelligence in Health Informatics, vol. 12, no. 3, pp. 120128, 2021, doi: 10.1016/j.jaihi.2021.120328.

7. Y. Zhang, H. Liu, and J. Zhao, "Machine Learning Approaches for Food Ingredient Classification," in Proc. of the ACM Symposium on Computational Food Science (CFoodSci), San Francisco, CA, USA, 2021, pp. 7582, doi: 10.1145/CFoodSci.2021.00012.

8. M. A. Gonzalez, P. Fernandez, and R. Lopez, "Nutritional Information Extraction from Food Labels Using AI and OCR," IEEE Trans. on Computational Food Analytics, vol. 4, no. 1, pp. 3038, 2022, doi: 10.1109/TCFA.2022.9876543.

9. L. Zhang and W. Chen, "Food Ingredient Analysis and Health Risk Prediction Using Deep Learning," in Proc. of the IEEE International Conference on Deep Learning in Healthcare (DLH), Shanghai, China, 2022, pp. 112119, doi: 10.1109/DLH.2022.00025.

10. A. Sharma, R. Patel, and N. Sinha, "AI-Powered Food Label Analysis for Allergen Detection and Consumer Guidance," in Proc. of the International Conference on AI in Nutrition and Health (AINH), London, UK, 2023, pp. 5865, doi: 10.1109/AINH.2023.00009.

[11] J. E. Nalavade and T. S. Murugan, THRFuzzy: Tangential holoentropy-enabled rough fuzzy classifier to classification of evolving data streams, Journal of Central South University, vol. 24, 2017.

[12] J. E. Nalavade and T. S. Murugan, HRNeuro-fuzzy: Adapting neuro- fuzzy classifier for recurring concept drift of evolving data streams using rough set theory and holoentropy, Journal of King Saud University – Information Sciences, vol. 30, 2018.

[13] J. E. Nalavade and T. S. Murugan, HRFuzzy: Holoentropy-enabled rough fuzzy classifier for evolving data streams, International Journal of Knowledge-based and Intelligent Engineering Systems, vol. 20, 2016.

Published by : http://www.ijert.org

International Journal of Engineering Research & Technology (IJERT)

ISSN: 2278-0181

Vol. 14 Issue 05, May-2025

1. J. E. Nalavade, M. Gavali, N. Gohil, and S. Jamale, Impelling Heart Attack Prediction System using Data Mining and Artificial Neural Network, International Journal of Current Engineering and Technology, vol. 4, no. 3, pp. 1-5, Jun. 2014.

2. A. Kumar, J. E. Nalavade, V. Yeola, V. Vivek, and Y. Srivastava, An adaptive learning system based on ant colony algorithm, Journal of Soft Computing and Engineering (IJSCE), vol. 3, no. 2, pp. 212- 214, 2013

3. K. M. Matrouk, J. E. Nalavade, S. Alhasen, M. Chavan, and N. Verma, MapReduce framework based sequential association rule mining with deep learning enabled classification in retail scenario, Cybernetics and Systems, pp. 1-23, Jan. 2023, Taylor & Francis.

4. J. E. Nalavade, A. Patil, A. Buchade, and N. Jadhav, Deep Neural Network and GAN-Based Reversible Data Hiding in Encrypted Images: A Privacy-Preserving Approach, SN Computer Science, vol. 5, no. 1, p. 45, Nov. 2023.

5. C. Nashte, J. E. Nalavade, A. Darvemula, S. Singh, and M. Bhowmick, Differentiation of Malignant and Benign Breast Lesions Using Machine Learning Algorithms, Differentiation, vol. 4, no. 5, May 2017.

6. D. R. Kale, J. E. Nalavade, S. Hirve, S. Tamboli, P. S. Randive, and

   N. Karlekar, Quantum Machine Learning Algorithms for Optimization Problems: Theory, Implementation, and Applications, International Journal Of Intelligent Systems And Applications In Engineering, vol. 12, no. 4, Jun. 2024.

7. S. Hirve, D. R. Kale, J. E. Nalavade, and P. S. Randive, Artificial Intelligence in Sustainable Agriculture: Enhancing Efficiency and Reducing Environmental Impact, Industrial Engineering Journal, vol. 53, no. 05, no. 09, pp. 103-109, May 2024.

8. G. B. Yadav, J. E. Nalavade, K. Vayadande, A. M. Kulkarni, and S. Umbare, Randomly Generating Music Using Context-Free Grammars, Indian Journal of Technical Education, vol. 47, no. 01, p. 127, 2024.

9. J. E. Nalavade, A. N. Jadhav, S. B. Chavan, A. R. Patil, and S. D. Khatawkar, Neurological Disorder Classification using Convolutional Neural Networks, International Journal Of Intelligent Systems And Applications In Engineering, vol. 12, no. 19, pp. 512- 51, 2024.

10. J. E. Nalavade, A. N. Jadhav, S. B. Chavan, A. R. Patil, and S. D. Khatawkar, Neurological Disorder Classification using Convolutional Neural Networks, International Journal Of Intelligent Systems And Applications In Engineering, vol. 12, no. 19, pp. 512- 517, 2024.

11. J. E. Nalavade and A. Patil, Reversible Data Hiding in Encrypted Images using Deep Neural Network and GAN Model, in 2022 3rd International Conference on Computing, Analytics and Networks (ICAN), Nov. 2022, pp. 1-4.

12. J. E. Nalavade and R. S. Patil, Driver Drowsiness Detection System Using Deep Neural Network, in 2022 3rd International Conference on Computing, Analytics and Networks (ICAN), Nov. 2022, pp. 1-4.

13. D. R. Kale, A. Buchade, J. E. Nalavade, S. G. Sapate, and A. J. Umbarkar, Detecting Violations of Conditional Functional Dependencies in Distributed Database, in 2023 IEEE International Conference on Blockchain and Distributed Systems Security (ICBDS), Oct. 2023, pp. 1-4.