ML Food Survey Classification

This project explores survey responses related to three food items — Pizza, Shawarma, and Sushi to build a custom multi-class classification model.

Built With

---

Dataset

Survey questions included in questions.md

• Complexity rating (1 to 5)
• Expected number of ingredients (free-form text)
• Expected settings for serving (multi-select)
• Expected price (free-form text)
• Movie association (free-form text)
• Drink pairing (free-form text)
• People association (multi-select)
• Hot sauce preference (multi-select)

Exact survey questions are available # todo

---

Final Model Highlights

• Accuracy: ~88.4%
• Trained using 100 independent runs
• One-hot and BOW encoding
• Softmax activation
• L2 Regularization: 0.005
• Learning Rate: 0.005
• 2000 iterations

---

Model Features

• Q1: Complexity (numeric)
• Q2: Parsed ingredient count (numeric)
• Q3: One-hot encoding of selected settings
• Q4: Parsed price (numeric)
• Q5: Bag-of-words representation of associated movies
• Q6: Drink category (manually curated → one-hot encoding)
• Q7: One-hot encoding of associated people
• Q8: Hot sauce preference (mapped from text to scale 0–4)

---

Evaluation & Results

Accuracy Results (100 Runs)

• Training: 91.2%
• Validation: 88.9%
• Test: 88.4%

Figures

• Accuracy stripplot: test_accuracy_stripplot.png
• Heatmaps and bar charts for various features (drinks, people, movies)

---

Getting Started

1. Clone the repo:

git clone https://github.com/lcai62/ml-food-classification/
cd ml-food-classification

2. Install requirements

pip install -r requirements.txt

3. Train model

python3 generate.py

4. Make predictions

python3 pred.py