This was one of Kaggle's competition; Expedia Hotel Recommandations
It's written for Python 2.7.2 and it's based on 'scikit-learn'
It was made to improve recommendation algorithm of Expedia hotel recommendation system. Expedia wanted to predict which hotel cluster user would belong to.
'Ensemble_hotel_cluster.py' is the result to solve the problem. It was very difficult to find proper model and columns combinations. In addition, Prediction Accuracy (0.30) is too low to judge it works.
However, train dataset has some specific regularities between them;
- user_location_country: The ID of the country the customer is located
- hotel_country: Hotel country
- srch_destination_id: ID of the destination where the hotel search was performed
- nights: How long they hope to stay in hotel
- prepare: How much time they are prepared beforehand
Finally, the target was changed 'hotel cluster' to 'is_package'.
'is_package' can be predicted by DecisionTreeClassifier at 90%.
The code file is 'DTC_find_package.py'
The train file, however, is too big to upload github. I leave a link where you can download the train file. There are four files on the web site and only one of them used.
- train.csv
link : https://www.kaggle.com/c/expedia-hotel-recommendations/data