🏡 Predicting Airbnb Listing Success

This project builds a full end-to-end machine learning pipeline to predict the success tier of an Airbnb listing — categorized as Excellent, Good, or Average — using only publicly available metadata.

🔎 Goal: Help real estate agents, analysts, and hosts evaluate listing potential before deployment, based on observable features like price, reviews, availability, and host trust signals.

---

🎯 Project Summary

• Type: Multi-class classification
• Target: success_label (Excellent / Good / Average)
• Model: Tuned Random Forest
• Evaluation: Accuracy, F1 Score, Confusion Matrix, Feature Importance

---

🧠 What Defines Success?

Since Airbnb doesn’t provide a built-in success score, we engineered a label using:

• ✅ adj_rating: Adjusted guest rating (normalized scale)
• ✅ log_reviews_ltm: Log-transformed review count (last 12 months)
• ✅ has_rating: Has the listing ever been rated?
• ✅ instant_bookable: Can it be booked without host approval?
• ✅ is_superhost: Airbnb’s trust-based badge

These features were combined into a composite success score, and quantile binned into 3 classes to ensure balance.

---

🧩 Features Used

Cleaned and engineered features from 5 thematic blocks:

Block	Examples
📍 Location	latitude, longitude, neighbourhood_encoded
💵 Price	log_price, availability_365, instant_book_int
🧽 Reviews	has_rating, log_reviews_ltm, adj_rating
🏘️ Property	property_type_encoded, beds_per_guest, baths_per_guest
🧑 Host Trust	hrt_ord (ordinal host response time), is_superhost

Also:

• Target leakage was avoided through explicit column dropping
• Safe encoding strategies like train-only target encoding were applied

---

📊 Modeling Approach

• Baseline: Random Forest (n=100, default)
• Tuned Model: GridSearchCV over max_depth, min_samples_split, n_estimators
• Scoring: f1_weighted (to account for class imbalance)

Key Metrics

Metric	Tuned Model
Accuracy	~0.71
Weighted F1 Score	~0.71
Macro F1 Score	~0.68
Avg Tree Depth	~30–40

---

📈 Feature Importance (Top 7)

1. log_reviews_ltm
2. has_reviews
3. latitude, longitude
4. log_price
5. availability_365
6. hrt_ord
7. neighbourhood_encoded

---

🔒 Clean Modeling Practices

• 🚫 No leakage: all features used in target creation were excluded from training
• 🧪 Stratified train-test split
• 🧼 Final datasets (train_df_ready.csv, test_df_ready.csv) exported cleanly

---

📁 Files in this Repository

File	Description
airbnb_success_pipeline.ipynb	Full project notebook
train_df_ready.csv	Cleaned training data
test_df_ready.csv	Cleaned test data
README.md	This file

---

📂 Download Processed Datasets

Due to GitHub's file size limitations, the full processed CSVs are hosted externally:

• 🔗 Download csv files here

To use them:

1. Download all files
2. Place them in the root of this project directory
3. Open airbnb_success_prediction.ipynb and run the notebook

---

👨‍💻 Author

Shawn Waringu
Data Scientist & Analyst

LinkedIn GitHub

---