Cuisine Classification Project

Project Overview

This project focuses on building a machine learning pipeline to classify cuisines based on their ingredients. Using a dataset of recipes, the goal is to predict the type of cuisine (e.g., Italian, Indian, Chinese) given a list of ingredients. The project explores various machine learning algorithms, feature engineering techniques, and dimensionality reduction methods to achieve optimal classification performance.

---

Dataset Description

The dataset is provided in the JSON file: train.json.

Dataset (train.json)

• Contains three columns:
  • id: Unique identifier for each recipe.
  • cuisine: The type of cuisine (target variable).
  • ingredients: A list of ingredients for each recipe.

---

Project Workflow

1. Data Preprocessing

• Cleaned the ingredients by:
  • Removing descriptive modifiers (e.g., "chopped onions" → "onions").
  • Standardizing ingredient names using mappings (e.g., "kosher salt" → "salt").
• Combined ingredients into a single string, separated by |||, for text vectorization.
• Split the train.json file into training and test sets (since the provided test.json lacks labels).

2. Feature Engineering

• Applied TF-IDF Vectorization to convert the list of ingredients into numerical vectors suitable for machine learning models.
• Reduced the dimensionality of the TF-IDF vectors using Principal Component Analysis (PCA):
  • Retained 90% of the variance by reducing to 1468 components.

3. Clustering

• Performed K-Means Clustering on the PCA-reduced data to group cuisines into clusters.
• Used the cluster labels as additional features for supervised classification models.

4. Machine Learning Models

Trained and evaluated the following models:

• K-Nearest Neighbors (KNN)
• Naïve Bayes (NB)
• Logistic Regression (LR)
• Random Forest (RF)
• Decision Tree (DT)
• Support Vector Machine (SVM)
• Stochastic Gradient Descent (SGD)

5. Hyperparameter Tuning

• Optimized model performance using GridSearchCV for hyperparameter tuning.
• For example, tuned C, gamma, and kernel for SVM.

---

Results

Model Comparison

Model	Accuracy
Support Vector Machine (SVM)	78.75%
Logistic Regression	77.28%
Stochastic Gradient Descent (SGD)	76.89%
Random Forest	64.49%
K-Nearest Neighbors	55.93%
Decision Tree	48.37%

Clustering Visualization

• Performed 2D visualization of cuisines in PCA-reduced space with cluster labels from K-Means.
• Provided insights into how cuisines group based on their ingredients.