Description: This project aims to create a Nutri-Score calculator for food products available in the United Kingdom. Nutri-Score is a nutrition label that provides at-a-glance information about the nutritional quality of a food product. The purpose of this project is to help consumers make healthier food choices by categorizing products and calculating their Nutri-Score based on various nutritional attributes. By leveraging machine learning techniques, this tool offers an interactive interface for users to search and evaluate different food products, promoting informed dietary decisions.
Introduction: In today's fast-paced world, making healthy food choices is crucial for maintaining overall well-being. However, understanding the nutritional content of various food products can be challenging for consumers. The Nutri-Score calculator project addresses this challenge by providing an intuitive platform for users in the UK to assess the nutritional quality of food items effortlessly. By categorizing products and calculating Nutri-Score based on key nutritional factors, this project empowers individuals to make informed decisions about their diet, promoting healthier lifestyles and contributing to improved public health.
Technologies Used: Python: The project is implemented using the Python programming language. Pandas: Pandas is used for data manipulation and analysis, facilitating the handling of the dataset. scikit-learn: The scikit-learn library is utilized for machine learning tasks, including imputation and regression modeling. HistGradientBoostingRegressor: This regressor from scikit-learn is employed for model-based imputation of missing values in the dataset. LabelEncoder: LabelEncoder from scikit-learn is used to encode categorical variables into numerical format. openpyxl: The openpyxl engine is utilized to read Excel files, enabling seamless data ingestion. This project showcases the intersection of data science and nutrition, offering a practical solution to enhance public awareness about the nutritional content of food products. Through the application of machine learning algorithms and interactive user interfaces, it bridges the gap between complex nutritional data and everyday consumers, encouraging healthier dietary choices in the United Kingdom.
Project Structure: The project directory is organized as follows:
/data: Contains the dataset files, including Final_UK_Food_Product.xlsx, which stores the raw data. /src: This directory houses the Python source code files. data_preparation.py: Handles data sourcing, preprocessing, missing value imputation, and feature engineering. nutri_score_calculator.py: Implements the Nutri-Score calculation logic. user_interaction.py: Manages user input processing and interaction. /output: Stores the output files or processed datasets. README.md: Provides project documentation and instructions for users and developers.
Data Preparation: Data Sourcing: The dataset, Final_UK_Food_Product.xlsx, was sourced from a reliable nutrition database, providing detailed information about various food products available in the United Kingdom.
Data Preprocessing Steps:
Filtering Data: Rows containing products from the United Kingdom were filtered using the 'countries_en' column. Label Encoding: Non-numeric columns such as 'main_category' were label-encoded for model input. Handling Missing Values: Missing values in selected nutritional columns were imputed using the HistGradientBoostingRegressor model. Data was split into features and target variables for training and prediction. Feature Engineering: The 'nutri_score' was calculated based on a weighted combination of nutritional attributes, representing the overall nutritional quality of each product.
Machine Learning Model: Algorithm: HistGradientBoostingRegressor Algorithm Overview: HistGradientBoostingRegressor is an implementation of gradient boosting machines using histogram-based learning. It's specifically designed for large datasets, making it efficient for both memory usage and computation time. This algorithm builds decision trees sequentially, with each tree correcting the errors made by the previous ones, ultimately improving the model's accuracy.
Key Features of HistGradientBoostingRegressor:
Histogram-Based Learning: Utilizes histograms to bucketize the input features, reducing the memory usage and speeding up the training process, especially for large datasets. Gradient Boosting: Builds an ensemble of decision trees, where each subsequent tree focuses on correcting the errors made by the combined predictions of the previous trees. Handling Missing Values: Supports missing values in the input data, making it suitable for imputing missing values in datasets. Regularization: Incorporates L2 regularization, preventing overfitting by penalizing large coefficients in the decision trees. Early Stopping: Allows early stopping based on a validation dataset to prevent overfitting and find the optimal number of boosting rounds. Hyperparameter Tuning: In a real-world application, fine-tuning the hyperparameters of the HistGradientBoostingRegressor can significantly impact the model's performance. Here are some key hyperparameters that could be tuned:
learning_rate (or eta): Controls the contribution of each tree to the final prediction. Lower values require adding more trees to maintain model performance. Typical values range from 0.01 to 0.3.
max_iter (or n_iter_no_change): Maximum number of boosting iterations (trees) to be run. Setting an appropriate value is crucial to avoid overfitting. It can be determined through cross-validation.
max_depth: Maximum depth of the individual decision trees. Deeper trees can capture more complex patterns but might lead to overfitting. It's essential to find a balance based on the dataset complexity.
min_samples_leaf: Minimum number of samples required to be at a leaf node. Increasing this value can prevent the model from being too specific to the training data.
l2_regularization: Strength of L2 regularization on leaf weights. Higher values penalize large coefficients, preventing overfitting. Common values are in the range of 0.0 to 1.0.
max_bins: Maximum number of bins to bucketize the input features. Increasing this value can improve model accuracy but also requires more memory. It's essential to monitor memory usage, especially for large datasets.
early_stopping: Enable early stopping based on a validation dataset to prevent overfitting. It stops training when the validation score doesn't improve for consecutive iterations (n_iter_no_change).
In practice, techniques like Grid Search or Randomized Search can be employed to explore combinations of these hyperparameters and find the optimal configuration for the given dataset. Cross-validation is crucial to validate the model's performance across different subsets of the data during hyperparameter tuning.
Hyperparameter Tuning: The model's hyperparameters were set to default values for simplicity in this project. However, for a real-world application, hyperparameters could be fine-tuned using techniques like grid search or randomized search to optimize the model's performance.
User Interaction: User Input Processing: The user can input a product name or category of interest.
Example 1: User inputs 'chocolate'. The system searches and displays chocolate products and their Nutri-Score ratings. Example 2: User inputs 'snacks'. The system filters snacks products and presents their Nutri-Score information. Expected Outcomes:
The system returns a list of products matching the user's input, along with their respective Nutri-Score ratings. Users can make informed decisions based on the Nutri-Score and product categories, promoting healthier food choices. This interactive feature enhances user engagement, making it easier for consumers to access nutritional information and choose healthier food options.