TravelTide – Customer Segmentation and Rewards Program Analysis

Project description

This project analyzes user behavior at TravelTide, an e-booking startup, to help the marketing team design a personalized rewards program.
The goal is to identify distinct customer personas and assign them perks they are most likely to value, improving retention and engagement.

The analysis combines SQL, Python, and machine learning to uncover behavioral patterns, validate marketing personas, and suggest data-driven campaign strategies.

---

Project summary

🎯 Objective

Support TravelTide’s Marketing Team in building a personalized rewards program by:

1. Understanding customer behavior and preferences.
2. Identifying behavioral segments (personas).
3. Assigning each customer a likely favorite perk for targeted marketing.

---

🧭 Approach

The project follows a structured four-stage workflow:

1. Exploratory data analysis (EDA) – Cleaned and explored customer, session, flight, and hotel data from PostgreSQL.
2. Feature engineering – Created user-level behavioral and value metrics (e.g., RFM, discount use, travel patterns).
3. Customer segmentation – Combined descriptive statistics (RFM + personas) with unsupervised ML (K-Means, DBSCAN).
4. Supervised modeling – Built Random Forest and Decision Tree classifiers to validate and operationalize personas.

---

💡 Key insights

• Travel behavior varies along continuous dimensions — customers rarely fit into rigid clusters.
• Descriptive segmentation (RFM + personas) gives greater interpretability than K-Means alone.
• Supervised ML validated that manual personas reflect measurable differences in user behavior.
• The decision tree revealed interpretable, high-level rules that can directly guide marketing segmentation.
• Iterating between manual persona definition → ML validation → rule extraction enables continuous improvement.

---

📈 Outcomes

• Around 6,000 user profiles aggregated and cleaned
• 16 engineered features including recency, frequency, and monetary value
• 5 RFM-based segments mapped to 8 behavioral personas
• Random Forest accuracy: 93%
• Decision Tree accuracy: 70%, offering clear, human-readable segmentation rules

---

📂 Links to key files

• Executive Summary (PDF)
• EDA Notebook
• Aggregated Analysis and Feature Engineering Notebook
• Machine Learning Notebook
• User Aggregated Dataset
• User Personas and Perks

---

🧱 Directory structure

TravelTide/
├── data/
│   └── traveltide_user_agg_features.csv
├── notebooks/
│   ├── TravelTide_-_EDA.ipynb
│   ├── TravelTide_-_Aggregated_Analysis_and_Feature_Engineering.ipynb
│   └── TravelTide_-_Machine_Learning.ipynb
├── reports/
│   └── Aggregated Analysis and Feature Engineering/
├── scripts/
└── README.md

---

⚙️ Installation

git clone https://github.com/<yourusername>/TravelTide.git
cd TravelTide
pip install -r requirements.txt

---

▶️ Usage

Run each notebook in order to reproduce the analysis:

1. TravelTide_-_EDA.ipynb
2. TravelTide_-_Aggregated_Analysis_and_Feature_Engineering.ipynb
3. TravelTide_-_Machine_Learning.ipynb

---

🧩 Dependencies

• pandas
• numpy
• matplotlib
• seaborn
• scikit-learn
• scipy
• sqlalchemy / psycopg2 (for PostgreSQL connection)

---

🗝️ Example usage

import pandas as pd

# Load aggregated user-level data
df = pd.read_csv('data/traveltide_user_agg_features.csv')

# Display key metrics
df[['ltv_usd', 'recency_days', 'frequency_trips_per_day']].describe()

---

🧠 Author

Pino Bonetti Marketing strategist & data analyst — Berlin Project completed as part of the Masterschool Data Analytics Program