[EN] Personalized Nutrition Assistant: Data-Driven Recipe Recommendations

📖 Background

With the rise of health awareness, people are increasingly seeking personalized dietary recommendations to meet their nutritional goals. However, creating a system that provides accurate, user-specific recipe suggestions presents several challenges:

1. Data Diversity: Nutritional datasets often vary in format and quality, making it hard to standardize for analysis.
2. User-Specific Needs: Different users have unique dietary goals (e.g., weight loss, muscle gain, or managing health conditions), requiring tailored recommendations.
3. Model Interpretability: Users need to understand why certain recipes are recommended, which demands a balance between accuracy and explainability.

To address these challenges, I developed the Personalized Nutrition Assistant, a system that leverages data science and machine learning to provide tailored recipe recommendations, helping users achieve their health goals.

---

✨ Our Solution

1. Data Collection and Cleaning

Background

The foundation of any recommendation system is high-quality data. I sourced a dataset of 1,174 food items from Kaggle, which included nutritional information (calories, protein, fats, etc.) and user inputs (dietary preferences and goals). However, the raw data was inconsistent, with duplicates and missing values.

• I used pandas to clean and standardize the dataset, removing duplicates, filling missing values, and normalizing nutritional data for consistency.
• I also integrated user inputs (e.g., daily calorie targets, preferred cuisines) to ensure the system could adapt to individual needs.

2. Recipe Recommendation Model

To provide accurate recipe recommendations, I employed the XGBoost model, a powerful gradient-boosting framework known for its performance in structured data tasks.

• I engineered features such as nutritional balance (protein-to-carb ratio), user preferences (e.g., vegetarian, low-carb), and goal alignment (e.g., weight loss).
• I trained the XGBoost model with a learning rate of 0.1 and a max depth of 5, achieving a training score of 0.62.
• The model predicts recipes that best match the user’s nutritional goals while considering their preferences.

3. User Interface with Streamlit

To make the system accessible, I developed a user-friendly interface using Streamlit, a Python framework for building interactive web apps.

• Users can input their dietary goals (e.g., target calories, preferred diet type) and receive instant recipe recommendations.
• The interface displays nutritional breakdowns and explains why each recipe was recommended, enhancing transparency.

---

🏆 Results & Achievements

Data Outcomes

• Dataset Size: Processed and standardized 1,174 food items from Kaggle, ensuring data consistency for training.
• Model Performance: The XGBoost model achieved a training score of 0.62, providing reliable recipe predictions.

Qualitative Analysis

• Personalization: The system successfully tailors recommendations to user goals, such as suggesting high-protein recipes for muscle gain or low-calorie options for weight loss.
• User Experience: The Streamlit interface offers an intuitive experience, with clear nutritional insights and explainable recommendations.

---

📲 App Development

The Personalized Nutrition Assistant is deployed as a web app using Streamlit, allowing users to access it from any device with a browser. The app integrates the XGBoost model and provides a seamless user experience.

Core Features

1. User Input Form: Users can specify their dietary goals, preferences, and restrictions (e.g., vegetarian, gluten-free).
2. Recipe Recommendations: The app displays a list of recommended recipes with nutritional details.
3. Explainability: Each recommendation includes a breakdown of how it aligns with the user’s goals.

App Screenshots

User input form for dietary goals	Recommended recipes with nutritional breakdown

📷 Demo

---

🚀 Benefits & Real-World Application

Achievements & Advantages

• Data-Driven Recommendations: The system uses cleaned, standardized data to provide accurate, personalized recipe suggestions.
• User-Centric Design: The Streamlit interface ensures accessibility and transparency, enhancing user trust.
• Scalable Model: XGBoost’s efficiency allows the system to handle larger datasets as user demand grows.

Practical Application

1. Health and Wellness: Helps users achieve dietary goals, such as weight management or improved nutrition.
2. Educational Tool: Can be used by nutritionists or educators to teach about balanced diets.
3. Commercial Use: Potential for integration into fitness apps or meal planning services.

---

🎯 Future Prospects

1. Model Enhancement: Incorporate more advanced models (e.g., neural networks) to improve prediction accuracy.
2. Expanded Dataset: Include more diverse food items and cultural cuisines to cater to a global audience.
3. Mobile App Development: Develop a native mobile app for iOS and Android to enhance accessibility.

---

🤝 Summary

The Personalized Nutrition Assistant successfully delivers data-driven recipe recommendations, empowering users to make healthier dietary choices. I look forward to collaborating with health tech companies to expand this solution and further advance AI in nutrition and wellness.

---

[中文] 個人化營養助手：資料驅動的飲食建議系統

📖 背景

隨著健康意識的提升，越來越多人希望獲得個人化的飲食建議，以達成他們的營養目標。然而，開發一個能提供精準、用戶專屬飲食建議的系統，面臨以下挑戰：

1. 資料多樣性：營養資料集的格式與品質不一，難以標準化進行分析。
2. 用戶特定需求：不同用戶有獨特的飲食目標（例如減重、增肌或管理健康狀況），需要量身定制的建議。
3. 模型可解釋性：用戶需要了解為何推薦某些食譜，這需要在準確性與可解釋性之間取得平衡。

為了應對這些挑戰，我開發了 個人化營養助手，一個利用資料科學與機器學習提供個人化飲食建議的系統，幫助用戶實現健康目標。

---

✨ 我們的解決方案

1. 資料收集與清理

背景

任何推薦系統的基礎都是高品質的資料。我從 Kaggle 獲取了包含 1,174 項食品 的資料集，涵蓋營養資訊（熱量、蛋白質、脂肪等）與用戶輸入（飲食偏好與目標）。然而，原始資料不一致，存在重複與缺失值。

• 我使用 pandas 清理與標準化資料集，去除重複項，填補缺失值，並將營養數據正規化以確保一致性。
• 我還整合了用戶輸入（例如每日熱量目標、偏好菜系），確保系統能適應個人需求。

2. 飲食建議模型

為了提供精準的飲食建議，我採用了 XGBoost 模型，這是一個高效的梯度提升框架，適合處理結構化資料任務。

• 我設計了特徵，包括營養平衡（蛋白質與碳水化合物比例）、用戶偏好（例如素食、低碳水）以及目標匹配（例如減重）。
• 我用學習率 0.1、最大深度 5 訓練 XGBoost 模型，達到 0.62 的訓練分數。
• 模型能預測最適合用戶營養目標的食譜，同時考慮他們的偏好。

3. 使用 Streamlit 打造用戶介面

為了讓系統更易用，我使用 Streamlit 開發了一個直觀的網頁介面，這是一個用於構建互動式應用的 Python 框架。

• 用戶可以輸入他們的飲食目標（例如目標熱量、偏好飲食類型），即時獲得食譜建議。
• 介面顯示營養細節，並解釋每個食譜的推薦原因，提升透明度。

---

🏆 成果與成就

數據結果

• 資料集規模：處理並標準化了 1,174 項食品，確保訓練數據的一致性。
• 模型表現：XGBoost 模型達到 0.62 的訓練分數，提供可靠的食譜預測。

質性分析

• 個人化：系統能根據用戶目標提供客製化建議，例如為增肌建議高蛋白食譜，或為減重建議低熱量選項。
• 用戶體驗：Streamlit 介面直觀，提供清晰的營養資訊與可解釋的建議。

---

📲 應用程式開發

個人化營養助手以 Streamlit 網頁應用形式部署，用戶可透過任何瀏覽器設備使用。應用整合了 XGBoost 模型，提供流暢的用戶體驗。

核心功能

1. 用戶輸入表單：用戶可指定飲食目標、偏好與限制（例如素食、無麩質）。
2. 食譜建議：應用顯示推薦食譜清單，包含營養細節。
3. 可解釋性：每個建議包含與用戶目標匹配的細節，提升信任。

應用截圖

用戶輸入飲食目標的表單	推薦食譜與營養細節

📷 影片演示

---

🚀 實際應用價值

成就與優勢

• 資料驅動建議：系統使用清理後的標準化資料，提供精準的個人化食譜建議。
• 用戶導向設計：Streamlit 介面確保易用性與透明度，提升用戶信任。
• 可擴展模型：XGBoost 的高效能讓系統能處理更大的資料集，滿足成長需求。

實際應用

1. 健康與養生：幫助用戶達成飲食目標，例如體重管理或改善營養。
2. 教育工具：可供營養師或教育者用於教授均衡飲食。
3. 商業應用：可整合進健身應用或餐飲規劃服務。

---

🎯 未來展望

1. 模型增強：引入更進階的模型（例如神經網絡），提升預測準確率。
2. 擴展資料集：納入更多樣化的食品與國際菜系，服務全球用戶。
3. 行動應用開發：開發 iOS 與 Android 原生應用，提升可達性。

---

🤝 總結

個人化營養助手成功實現了資料驅動的飲食建議，幫助用戶做出更健康的飲食選擇。我期待與健康科技公司合作，擴展此解決方案，進一步推動 AI 在營養與養生領域的應用。