🔗 GitHub repository

🔗 Dataset Source

Description

We use the Inside Airbnb dataset to identify significant trends in customer interest and predict the pricing of rentals in the US and Europe. This problem is similar to a classical use case of machine learning: house price prediction. We can then provide solutions to business difficulties. We have used LIME and SHAP to interpret the model’s predictions at the local and global levels.

Model Interpretation

  1. Linear Regression: - We use LIME and SHAP to interpret the model’s predictions at the local and global levels. We take a random test vector from the test set and set the number of the top features as 5, the number of perturbed samples as 5000. my notes

The comparatively higher market value of the house reflected by the provided vector (depicted by a bar on the left) can be related to the following reasons:

  • The high value of room_type_private_room on the negative side indicates that because it’s a private room, the price will reduce.
  • The high value of longitude on the negative side indicates that the house is not in a suitable place such as city centers or crowded neighborhoods.
  • The high value of bedrooms on the positive side indicates that the house has a bedroom, which is a private room and people prefer it.
  • The high value of review_scores_rating on the positive side indicates that the house has a high rating. my notes
  • At the global level, we can see that longitude, bedrooms, accommodates, latitude, beds, review_scores_rating affect the price most.

  1. Decision Tree Regression: - At the global level, we can see that longitude, accommodates, latitude, bedrooms affect the price most. my notes

  2. Random Forest Regression: - At the global level, we can see that longitude, accommodates, latitude, bedrooms affect the price most. my notes

  3. XG BOost: - At the global level, we can see that bedrooms, room_type_entire_home/apt, accommodates, longitude, latitude affect the price most. my notes

🔗 GitHub Code