Create Optimal Hotel Recommendations Data set: found at Kaggle:Expedia Hotel Recommendations& Train.csv Requirements: ·Write a one-page summary of your approach in getting to your prediction...

OPTIMAL HOTEL RECOMMENDATION.docx
OPTIMAL HOTEL RECOMMENDATION
The task was to predict the hotel cluster a user would book based on his/her booking history. The data given included user’s search trend, interaction with searches and booking history. Also, various features were also given corresponding to each hotel’s review.
The data provided for the problem had over 9 million data points with 24 feature vectors in train.csv and over 62 thousand data points with 150 feature vectors in destination.csv. Thus, the data available in hand was very big. The main task while choosing the correct algorithm was to manage this data. Also, the pre-processing steps showed a uniform distribution of hotel cluster choice thus dimensionality reduction was the main focus.
1. Data Preprocessing: We assumed that people who made more than 20 bookings were actually travel agents and thus didn’t represent the target population. Thus, they were removed from the data. People who didn’t make bookings and were click counts also weren’t a part of our target population and they also were removed. There were three columns which had date associated with them. This information as it is wasn’t useful, so we extracted month and year from them as additional features and the original features were dropped. After chi square test and PCA we came to conclusion that our current feature vector except customer id was a good final data.
2. Data Splitting: after combining the destination data through the common key, the data was split into a ratio of 3:1 in train and test set.
3. Modeling: We tried 4 different approaches toward model for the task. Each model’s accuracy was judged through the confusion matrix built upon the test data set. Since the data still had few abnormalities, we didn’t see exceptional performance by any model. The task could be called a multiclass classification problem. Through initial data preprocessing we found that...