High or Low? Predict Recipes Rating with Complexity & Nutritional Health Information
Introduction
This project uses the Recipes and Ratings dataset from Food.com, which combines two tables of user‑submitted recipes and their corresponding reviews originally scraped for a recommender‑systems study. And We are interested in exploring:
How do a recipe’s nutritional health information and complexity relate to its ratings?
Why it matters
Home cooks, nutritionists, and recipe‑platform designers should care because understanding these relationships can:
- Help users discover recipes they’re more likely to enjoy
- Support healthier choices by revealing how nutrition influences ratings
- Optimize recommendation engines for both taste and health
Preparing the data
- Merging We left merged the datasets to get all reviews of recipes included in the recipes dataset.
- Zero-ratings Change to NaN In reality, the lowest score a reviewer can actively give is 1. A recorded value of 0 almost always means no rating was actually provided even though the review text was submitted. We therefore convert rating = 0 to NaN before analysis so that missing opinions don’t bias averages.
- Averaging Rating We calculated the average rating per recipe and added it as a new column.
Dataset Size
- Number of unique recipes (rows): 234429
- Number of columns in the merged dataset: 17
Relevant Columns & Descriptions
| Column | Description |
|---|---|
| minutes | Preparation time in minutes |
| n_steps | Number of instruction steps in the recipe |
| n_ingredients | Total count of distinct ingredients used |
| tags | Food.com tags for recipe (which contains some health-related tags) |
| nutrition | List [calories (#), total fat (PDV), sugar (PDV), sodium (PDV), protein (PDV), saturated fat (PDV), carbohydrates (PDV)]; PDV = % daily value |
| avg_rating | Mean user rating (1–5), computed by averaging all nonzero ratings submitted |
These are the features we are interested to investigate. By understanding how preparation complexity and nutritional makeup jointly drive popularity, we aim to build a recommender that balances taste‑appeal with health considerations.