Price prediction and classifying in action!
The Price Isn't Right
Finding an Airbnb isn't an easy job. Especially if you're planning with others, going through tens of hundreds of listings before a trip can be exhausting.
The two biggest concerns we always have when we look for Airbnbs is the price and what type of listing it is. That's why we built "The Price Isn't Right"!
We've developed a chrome extension that automagically finds the current listing you're looking at and runs machine learning classifiers and clustering to give you the predicted price of a page and the classified listing type (with 5 types - "Modern", "Luxury", "Tourist", "Homey" and "Quiet").
Predictions and clustering is based on the Seattle 2016 dataset - so far it's going to be most accurate for Seattle Airbnbs!
A quick dip into how we made this:
Our chrome extension calls the Airbnb API every time someone is visiting a listing page and clicks our chrome extension.
We parse this data and feed this into our pickled classifiers.
Our price prediction is based on a few selected and engineered features from our dataset, fed in to a Gradient Boosted Regressor.
Our clustering also feature engineers data that we think would be indicative of a "cluster" of listings. Next we use a one-hot encoder to encode our features into a sparse matrix of one-hot vectors which we use feature decomposition (more specifically singular value decomposition) to further decrease the dimension of the dataset. Then we feed this into a Agglomerative Clustering algorithm to label our results (we used 5 clusters). Since this is a transductive learning algorithm, we need to feed these labels (and our original data) into a linear simple vector classifier in order to make a classifier that can make predictions based on the data we have.