Obesifix - Machine Learning

Hello, this is machine learning part of Obesifix application made by Capstone Team C23-PS344 ✨

Table of Contents

• Introduction
• ML Team
• What We Do?
• What We Use?
• Repositories
• Image Classification Model
• Recommendation System
• Machine Learning Model

Machine Learning Team

Name	Bangkit ID	Contacts
Natasha Clarissa Maharani	M151DSY1486	Github & Linkedin
Melody Priscilla Tan	M151DSY1458	Github & Linkedin

What We Do?

We are developing a food classification & recommendation model that suggests suitable food options to users.

What Packages that we use in Google Colab/Jupyter Notebook?

Packages
Tensorflow
Keras
Pandas
Scikit-Learn
Numpy
Matplotlib

Repositories

Learning Paths	Link
Organization	Github
Machine Learning	Github
Machine Learning API	Github
Mobile Development	Github

Image Classification Model

Model	Classification Output
Transfer Learning : InceptionV3	Apple, Banana, Chicken Curry
Input : image(256, 256)	Chicken Wings, Donuts, French Fries
Output : 19 labels	Fried Chicken, Fried Rice, Hamburger
Total params: 23,851,784	Hot dogs, Ice Cream, Omelette
Trainable params: 23,817,352	Onion Rings, Orange, Pancakes, Pizza
Time : 5440s	Rice, Spaghetti, Sushi

Recommendation System

Using KNN we filter the food according to user's preference and health condition.

Food Preferences (19 type of preferences)
Beef, Bread, Breakfast, Chicken
Cookies, Dairy Product, Dessert, Duck
Fish, Fruit, Lamb, Lunch, Pasta
Pork, Rice, Seafood, Soup, Soy Product, Vegetable

Health Conditions
Underweight
Normal
Overweight
Obese

Machine Learning Model

The first model is for classifying food pictures into 19 categories in which we will be able to calculate the nutrition of this food. For the model, we built a sequential model using Tensorflow and Keras API. We use transfer learning InceptionV3. Our input are trained through some parts of the InceptionV3 layers. Then the output is flattened. After that, it is then passed into a Deep Neural Network with Dropout Layers.

For the second model we use KNN for product based recommendation. The recommendation is based on the health condition and preferences.