Data source : http://insideairbnb.com/ [9 Feb, 2021 London Archives] Content: Listings and Reviews
Neighbourhood_group column is 100% missing, hence removed. Last_review column’s missing values have been imputed by 0. Assumption: there are no reviews for those listings Name identifiers have been removed since we have ids if needed
INSIGHTS: Host 33889201 outmatches others by a lot (800+ listings) Westminster is the neighbourhood with the most listings Most listings are of the type: Entire home/apt or Private room
Price, no. of reviews and minimum nights are highly skewed For a better visualization, we remove extreme values of price during bivariate analysis.
Hotel rooms have the highest range in prices, closely followed by Entire home/apt. In general,Hotel rooms and Entire home/apt are more expensive than Shared or Private rooms Listings in central London have, in general, high availability Most reviews are in English language out of 44 total languages
Hotel rooms have the highest range in prices, closely followed by Entire home/apt. In general,Hotel rooms and Entire home/apt are more expensive than Shared or Private rooms Listings in central London have, in general, high availability Most reviews are in English language out of 44 total languages
Business Use Case : Ease of processing feedback from customers Methodology : Unsupervised learning English comments using Vader Created variable : Sentiment Score (Range: -1 to 1, measuring positive/negative/neutral sentiment) Bucketed Score as per industry standards to label sentiments as positive, medium, or negative.
Most comments are positive People like when the place is clean and comfortable the host is friendly and the neighbourhood is centrally located. Performance of most reviewed listing dropped in 2020 probably due to Covid Highest Scored Listing-99241. Highest Scored Host- The Portobello Room Note- Only considered listings, hosts with at least 50 reviews for point 4.
Business Use Case : Predict price for a new listing based on variables in multiple datasets Methodology : Supervised Learning (Regression Analysis) Feature engineered variable: Avg_Sentiment_Score_Per_Listing using Score from sentiment analysis Experimented with different models and picked best. Performed Hyperparameter Tuning to get best set of parameters
Accommodates Avg_Sentiment_Score Neighbourhood Room Type Calculated Host Listings Count Host Identity Verified Host superhost Instant Bookable Minimum Nights Reviews Per Month Availability 365
Random Forest Regressor Validation RMSE: 249 Train RMSE: 159