European restaurants: Classification, Hypotheses & Recommender

The project classifies restaurants by various features using XGBoost and scikit-learn models and gives content-based recommendations of European restaurants using Jaccard metric from SciPy. Also aggregates the data for the Tableau Public dashboard.

Libraries: numpy, pandas, SciPy, scikit-learn, xgboost, matplotlib

Table of contents

• Dataset
• Machine Learning problems
• The Dashboard

Dataset

• The dataset of TripAdvisor's data is from Kaggle.

Machine Learning problems

• The first ML problem is classification of restaurants. The given problem was solved by using a few classifiers using scikit-learn.

  • Classificaton of Italian restaurants by "Serves Alcohol" feature. The XGBoost model has the best accuracy score - 77%.
  • Classificaton of the UK restaurants by "Vegetarian Friendly" feature. The Random Forest model has the best accuracy score - 95%.
  • Classificaton of German restaurants by "price_level" feature. The Gradient Boosting model has the best accuracy score - 74%.

The result is in the restaurant_classification.ipynb notebook.

• The second ML problem is building a content-based recommender by restaurant features. The given problem was solved by using the SciPy's Jaccord metric in the restaurant_recommender.ipynb notebook.

• The third problem is hypothesis testing about relationship between average rating and some restaurant features (Spanish cuisine, Cheap Food, Vegetarian Friendly) for different countries. The given problem was solved by using Student's and Welch's t-tests from SciPy in the restaurant_hypothesis_testing.ipynb notebook.

The Dashboard

• The result is used in the Tableau public dashboard.