📊 Dynamic Pricing System for Short-Term Rentals

---

🚀 Project Overview

The Dynamic Pricing System is a machine learning project designed to generate data-driven nightly price recommendations for short-term rental listings (e.g., Airbnb).

Pricing rental properties effectively is a complex problem — setting prices too high reduces bookings, while pricing too low leads to revenue loss. This system leverages historical listing data and advanced feature engineering techniques to predict optimal pricing strategies.

💡 This project demonstrates an end-to-end ML pipeline — from data preprocessing to model tuning and evaluation — with a focus on real-world applicability.

---

Check it Out here

Dynamic-Pricing-System: https://dynamic-pricing-system-bbovjcwe6xyf5ujvnmkdao.streamlit.app

🎯 Problem Statement

Hosts often rely on intuition rather than data, ignoring key factors such as:

• Local market trends
• Property characteristics
• Listing availability
• Review activity
• Competition within neighbourhoods

This results in suboptimal pricing decisions.

👉 This project builds a predictive pricing engine that learns patterns from historical data to recommend competitive prices.

---

🧠 Key Highlights

• ✔️ End-to-end ML pipeline (EDA → Feature Engineering → Modeling → Evaluation)
• ✔️ Advanced feature engineering (log transforms, target encoding, clustering)
• ✔️ Handling of high-cardinality features (neighbourhood encoding)
• ✔️ Hyperparameter tuning using RandomizedSearchCV
• ✔️ Comparison of multiple models (Linear, KNN, Random Forest, Boosting)
• ✔️ Real-world dataset with 220K+ listings

---

🧩 Features & Engineering

🔹 Core Features

• Location → City, Neighbourhood (target encoded)
• Property → Room type
• Availability → Minimum nights, availability_365
• Reviews → Number of reviews
• Host → Listing count

---

🔹 Advanced Feature Engineering

• 📌 Log Transformations → Reduced skewness in numerical features
• 📌 Target Encoding → Neighbourhood pricing signal
• 📌 KMeans Clustering (Location Intelligence) → Grouped latitude/longitude into meaningful regions
• 📌 One-Hot Encoding → Low-cardinality features (city, room type, clusters)
• 📌 Outlier Handling → Removed extreme pricing and minimum night anomalies

🚀 These steps significantly improved model performance and stability.

---

🛠️ Tech Stack

• Python 3.11
• Pandas / NumPy – Data processing
• Matplotlib / Seaborn – Visualization
• Scikit-learn – ML models & pipelines
• XGBoost – Gradient boosting model
• Jupyter Notebook / Google Colab – Development

---

📈 Model Performance

After feature engineering and hyperparameter tuning:

Model	MAE	RMSE	R² Score
Random Forest ✅	0.346	0.466	0.595
XGBoost	0.360	0.478	0.574
Gradient Boosting	0.367	0.485	0.561

Also tested out other regression models with the scaled dataset

Model	MAE	RMSE	R² Score
Linear Regression	0.73	169.3	0.22
KNN.	0.670	142.67	0.344

📊 The Random Forest model achieved the best performance, explaining ~59% of price variance.

---

🔧 Machine Learning Pipeline

1. Data Cleaning

   • Removed invalid and extreme values
   • Handled missing data

2. Feature Engineering

   • Log transformation of skewed features
   • Target encoding for neighbourhood
   • KMeans clustering on geolocation
   • One-hot encoding for categorical variables

3. Train/Test Split

   • Strict separation to prevent data leakage

4. Scaling

   • Applied where required (linear models)

5. Model Training

   • Linear Regression (baseline)
   • KNN
   • Random Forest (best performer)
   • Gradient Boosting
   • XGBoost

6. Hyperparameter Tuning

   • RandomizedSearchCV for optimal performance

---

📊 Exploratory Data Analysis (EDA)

• Distribution analysis of pricing and features
• Detection of extreme outliers
• Correlation analysis with target variable
• Feature behavior across cities and room types

---

💡 Key Insights

• 📍 Location is the strongest pricing driver
• 🏠 Room type significantly impacts price
• 📉 Extreme values (outliers) degrade model performance
• 🔄 Log transformation improves model stability
• 🤖 Ensemble models outperform linear approaches

---

📌 Dataset

• Source: Kaggle – NYC Airbnb Open Data
• Size: 226,000+ listings
• Features: Location, availability, reviews, host info

---

⚡ Project Status

• Data cleaning & preprocessing
• Feature engineering
• Model training & evaluation
• Hyperparameter tuning
• Model stacking (planned)
• Deployment (Streamlit)
• Advanced feature engineering (ongoing)

---

🚀 Future Improvements

• 🔹 Model stacking / ensembling
• 🔹 Advanced location features (distance-based metrics)
• 🔹 Time-based features (seasonality)
• 🔹 Deployment using Streamlit
• 🔹 Integration with real-time pricing APIs

---

🧑‍💻 How to Run

git clone https://github.com/AryanSharma1017/Dynamic-Pricing-System.git
cd Dynamic-Pricing-System
pip install -r requirements.txt

Run the notebook in Jupyter or Colab.

---

📫 Author

Aryan Sharma

• GitHub: https://github.com/AryanSharma1017
• LinkedIn: https://www.linkedin.com/in/aryansharma007

---

⭐ Final Note

This project demonstrates not just model building, but practical machine learning skills including:

• Feature engineering
• Handling real-world messy data
• Model tuning and evaluation
• Avoiding data leakage

🚀 Built with the mindset of solving real-world pricing problems using machine learning.