pregoogle

A simple web app that uses ML to predict if a recipe is safe or not safe for pregnant woman.

Problem

1. According to the National Center for Health care Statistics (NCHS), in 2016, there were approximately 1.1 million fetal losses in the United States. Approximately 10% of those are directly attributed to infectious diseases.
2. Popular online recipe review sites have begun to highlight allergy/gluten restrictions on their recipes. However, they do not yet have a feature that speaks to the unique dietary restrictions of pregnant women.
3. It’s difficult to identify whether a recipe is pregnancy-safe or not from simply scanning the ingredient list because:
   • There are a large number of unsafe foods.
   • Certain foods are only unsafe based on style of preparation.

Solution

Websites could start placing icons beside recipes not recommended for pregnant women.

In this prototype, I created a search site that will scrape AllRecipes.com based on the recipe ID to determine if the food is safe. The presentation can be seen here:

https://docs.google.com/presentation/d/1r1hnA1MNYzTb8FjMDCmpkGqsSh0F2YmVTOS1Nki4gXY/edit#slide=id.p3

Implmentation overview

1. Use python to scrape reviews from allrecipes.com
2. Manually label a few of these reviews
3. Clean data by eliminating pronouns, punctuation, etc.
4. Transform string into token count
5. Transform toekn count into TF-IDF score vectors
6. Train on TF-IDF vectors with Naive Bayes
7. Other techniques applied for optimization

• Cross validation
• Grid search

Results