Food label

Abstract

The project aims to create an innovative food product search and prediction system that enables users to explore a comprehensive database of 900,000 food items. Each product is enriched with metadata, including its name, brand, country of origin, category, nutritional values, and health- related scores like PNNS and Nutri-Score. The primary objective is to predict the NOVA classification of food items, which categorizes foods based on their level of processing, ranging from minimally processed to ultra-processed. Accurate NOVA classification provides critical insights into the health implications of food products and aligns with the dietary preferences of health-conscious users. To achieve this, three machine learning models were developed and trained using datasets con-taining varying numbers of nutritional parameters—7, 8, and 44 nutrients. These models were compared for predictive accuracy to determine the optimal set of features for NOVA classification. The availability of actual NOVA classifications within the dataset enabled robust model evaluation and accuracy benchmarking. A highly functional search engine was also developed, allowing users to filter and retrieve products based on attributes like NOVA class, product name, brand, and category. This feature empowers users to make informed dietary choices by providing quick and targeted access to information aligned with their preferences and health goals. The project faced challenges in managing the massive dataset stored in a CSV file. To facilitate efficient querying and analysis, the data was migrated into a database using Logstash, selected for its simplicity and reliability in handling such tasks. The system architecture was meticulously designed as a three-tier structure. The frontend, built with React, provides a user-friendly interface for search and browsing. A Node.js server serves as the intermediary, handling user requests and orchestrating backend services. For computational tasks like NOVA prediction, a Flask-based Python server integrates with machine learning models, ensuring high accuracy and seamless operations. Elasticsearch was chosen as the backend database to power the search engine. Known for its scalability and high-performance search capabilities, Elasticsearch allows the system to handle real-time queries efficiently, delivering lightning-fast search results for users. This combination of technologies ensures the system remains robust, responsive, and user-centric. In summary, this project leverages advanced machine learning, efficient data management tools, and a well-integrated architecture to deliver an effective food product search and prediction system. It serves as a practical tool for users seeking health-oriented food choices, highlighting the intersection of data science, user-centric design, and health informatics.