Hybrid Diet Recommendation System: Integrating Food Engineering Constraints with K-Nearest Neighbors

This repository contains a hybrid machine learning and deterministic application designed to generate localized, macronutrient-optimized meal plans. By combining established nutritional engineering formulas with a vector-based recommendation engine, the system processes user basal metrics and filters regional food profiles to deliver precise, target-aligned dietary frameworks.

The dataset is curated and feature-engineered specifically using regional nutritional profiles, providing an authentic application of data science to localized food systems.

---

System Architecture

The application bypasses the limitations of purely statistical models by utilizing a two-phase hybrid pipeline:

1. Deterministic Rule Engine (Basal Metrics)

Before executing predictive or filtering algorithms, the system establishes individual biological targets using exact physiological formulas.

• Body Mass Index (BMI): Calculated to baseline user health metrics.
• Basal Metabolic Rate (BMR): Determined via the modern Mifflin-St Jeor equation, chosen for its validated accuracy in clinical and field settings.
• Total Daily Energy Expenditure (TDEE): Quantified by applying physical activity multipliers to the calculated BMR baseline.
• Caloric Targeting: Configured programmatically based on user objectives (e.g., a continuous deficit constraint of 500 kcal for mass reduction or a 300 kcal surplus for mass gain).

2. Machine Learning Recommendation Engine

Once target macronutrient constraints are set, the system passes these vectors to a K-Nearest Neighbors (KNN) algorithm. The model evaluates the mathematical distance between the target meal profile and the available food matrix to output the closest structural matches.

---

Data Engineering & Localization

A core feature of this project is the integration of localized food profiles, derived from the official Food Composition Table for Bangladesh. Standard open-source nutrition datasets often fail to reflect regional eating habits accurately, rendering general models impractical for local deployment.

Feature Engineering Details

To optimize the recommendation matrix, the raw food composition data was engineered with custom binary constraint tags:

• Meal Classification Flags: is_breakfast, is_lunch, is_dinner, and is_snack. These prevent the model from making culturally incompatible recommendations (e.g., suggesting heavy carbohydrate main courses for breakfast or traditional street snacks for lunch).
• Dietary Profiling Flags: is_vegetarian, is_diabetic_friendly (low glycemic index tracking), and is_high_protein. These act as hard constraints to filter the dataset before calculating vector distances.

---

Mathematical Formulation

The system utilizes the following formal equations outside the machine learning pipeline to establish the target constraints:

Body Mass Index

$$\text{BMI} = \frac{\text{Weight (kg)}}{\text{Height (m)}^2}$$

Basal Metabolic Rate (Mifflin-St Jeor)

$$\text{BMR}_{\text{Male}} = (10 \times \text{Weight}_{\text{kg}}) + (6.25 \times \text{Height}_{\text{cm}}) - (5 \times \text{Age}_{\text{years}}) + 5$$

$$\text{BMR}_{\text{Female}} = (10 \times \text{Weight}_{\text{kg}}) + (6.25 \times \text{Height}_{\text{cm}}) - (5 \times \text{Age}_{\text{years}}) - 161$$

Vector Distance Metric

The KNN model utilizes Cosine Similarity to isolate food recommendations. This metric evaluates the angular distance between vectors rather than absolute magnitudes, ensuring the ratio of protein, fat, and carbohydrates matches the nutritional target, regardless of portion sizes.

$$\text{Cosine Similarity} = \frac{\mathbf{A} \cdot \mathbf{B}}{|\mathbf{A}| |\mathbf{B}|}$$

---

Technology Stack

• Language: Python
• Framework: Streamlit (Deployment and Interactive Interface)
• Data Manipulation: Pandas, NumPy
• Machine Learning: Scikit-learn (NearestNeighbors unsupervised learner)

---

Installation & Deployment

To run the pipeline and interactive web interface locally, ensure you have Python installed, then follow these steps:

1. Clone the repository:

   git clone https://github.com/yourusername/hybrid-diet-ml-planner.git
   cd hybrid-diet-ml-planner

2. Install dependencies:

   pip install streamlit pandas scikit-learn openpyxl numpy

3. Run the application:

   streamlit run "AI diet.py"

   Or on Windows, simply double-click run.bat.

4. The application will launch in your default browser at http://localhost:8501.

---

Project Structure

├── AI diet.py                                  # Main application script
├── Expanded_Bangladeshi_Food_Database_ML.xlsx  # Curated regional food dataset
├── FCT_10_2_14_final_version.pdf               # Reference: Food Composition Table for Bangladesh
├── run.bat                                     # Windows quick-launch script
├── .gitignore                                  # Git ignore configuration
└── README.md                                   # Project documentation

---

License

This project is developed for academic and research purposes.