✈️ Flight Ticket Price Prediction

Predicting Indian domestic flight prices using machine learning, built as part of the Simplon Maghreb × Jobintech Data Analysis bootcamp.

---

📌 Objective

Build a regression model that predicts flight ticket prices with a MAE under 1,250 INR (~15€), based on 300,000+ booking records from Indian domestic airlines.

---

📁 Repository Structure

├── Airline_Ticket_Prices_EDA.ipynb   # Exploratory Data Analysis
├── tests_statistic.ipynb             # Statistical hypothesis testing
├── model.ipynb                       # Preprocessing, modeling & evaluation
├── load_model.py                     # Example script for loading & using the model
├── model_metadata.json               # Saved model metrics and feature info
├── Clean_Dataset.csv                 # Source dataset
└── Documentation/                    # Additional project documentation

💾 The trained model (.joblib) is hosted externally due to file size — see Model.joblib

---

📊 Dataset

Property	Value
Source	Indian domestic flight bookings
Records	300,000+
Target	price (INR)

Features:

Variable	Type	Description
airline	Categorical	Carrier name (AirAsia, Vistara, IndiGo, etc.)
source_city	Categorical	Departure city
destination_city	Categorical	Arrival city
departure_time	Ordinal	Time-of-day slot (Early Morning → Late Night)
arrival_time	Ordinal	Time-of-day slot (Early Morning → Late Night)
stops	Ordinal	zero / one / two_or_more
class	Ordinal	Economy / Business
duration	Continuous	Flight duration in hours
days_left	Continuous	Days between booking and departure
price	Continuous	Ticket price in INR (target)

---

🔍 Project Phases

1. Exploratory Data Analysis (Airline_Ticket_Prices_EDA.ipynb)

• Variable type classification (continuous, discrete categorical, ordinal)
• Missing value and duplicate checks
• Univariate analysis: histograms, boxplots, and bar charts for all variables
• Multivariate analysis: relationships between each feature and price

Price distribution — bimodal structure reflecting Economy vs Business classes:

Airline vs average price — Vistara and Air India occupy the premium segment:

Stops vs average price — more stops correlates with higher price:

Days before flight vs price — prices spike sharply within 5 days of departure:

Average price by airline and number of stops — Vistara commands a consistent premium across all stop configurations:

Price by departure time and class — ticket class is the dominant price driver, departure time has marginal effect:

Key insights:

• class is the dominant price driver — Business tickets average ~8x more than Economy
• Booking within 5 days of departure causes a sharp price spike (~30,000 INR peak)
• Vistara is the most expensive airline across all stop counts
• More stops generally correlates with higher prices, likely due to premium airline routing

2. Statistical Hypothesis Testing (tests_statistic.ipynb)

Test	Variables	Result
Pearson correlation	duration vs price	r = 0.20, p ≈ 0 → significant but weak correlation
ANOVA	airline vs price	F = 17,194, p ≈ 0 → significant price differences across airlines
ANOVA	stops vs price	F = 6,477, p = 0 → significant price differences across stop counts
Chi-square (independence)	departure_time vs stops	p ≈ 0 → the two variables are dependent
Chi-square (goodness of fit)	AirAsia market share	Rejects H0 of 70% share — actual share is significantly different

3. Modeling (model.ipynb)

Preprocessing:

• Ordinal encoding: stops (0/1/2), class (0/1)
• OneHot encoding: airline, source_city, destination_city, arrival_time, departure_time
• Standard scaling: duration, days_left
• Train/test split: 80/20

Model comparison (5-fold cross-validation on training set):

Model	Selected
Baseline (DummyRegressor)	❌
Linear Regression	❌
Ridge Regression	❌
Random Forest	✅

Final model — Random Forest (n_estimators=50):

Metric	Value
R²	0.9857
MAE	1,063 INR (~12.7€) ✅
RMSE	2,722 INR
95% Confidence Interval	[1,043 — 1,083 INR]

✅ Business objective achieved: MAE < 1,250 INR (equivalent to ~15€)

---

🛠️ Tech Stack

• Python 3.14
• pandas / NumPy — data manipulation
• Matplotlib / Seaborn — visualization
• scikit-learn — preprocessing, modeling, evaluation
• SciPy — statistical tests
• joblib — model serialization
• uv — package management

---

🚀 Usage

Load the model and predict

import joblib
import pandas as pd
import numpy as np

pipeline = joblib.load("flight_price_pipeline.joblib")
encoder = pipeline["encoder"]
scaler  = pipeline["scaler"]
model   = pipeline["model"]

new_flight = {
    'airline': ['Vistara'],
    'source_city': ['Delhi'],
    'destination_city': ['Mumbai'],
    'departure_time': ['Evening'],
    'arrival_time': ['Night'],
    'stops': [1],          # 0=zero, 1=one, 2=two_or_more
    'class': [0],          # 0=Economy, 1=Business
    'duration': [2.5],
    'days_left': [15]
}

df_input = pd.DataFrame(new_flight)

categorical_cols = ['airline', 'source_city', 'destination_city', 'arrival_time', 'departure_time']
numerical_cols   = ['duration', 'days_left']
ordinal_cols     = ['class', 'stops']

encoded = encoder.transform(df_input[categorical_cols])
scaled  = scaler.transform(df_input[numerical_cols])

X_input = np.concatenate([df_input[ordinal_cols].values, scaled, encoded], axis=1)

predicted_price = model.predict(X_input)
print(f"Predicted price: {predicted_price[0]:,.0f} INR")

---

👥 Team

Built by the Data-Analysis-Hub team as part of the Simplon Maghreb × Jobintech Data Analyst training program (Cohort 2026).