Flight Price Prediction Using Machine Learning

Aman Preet Gulati 19 Feb, 2024 • 11 min read

Introduction

In this article, we will be analyzing flight fare prediction using a machine learning dataset using essential exploratory data analysis techniques then will draw some predictions about the price of the flight based on some features such as what type of airline it is, what is the arrival time, what is the departure time, what is the duration of the flight, source, destination and more.

Key Takeaways

  1. Learn the complete process of EDA (using a machine learning dataset)
  2. Learn to withdraw some insights from the dataset both mathematically and visualize it.
  3. Visualising the data to get better insight from it.
  4. We will also see what kind of stuff we can do in the feature engineering part.

This article was published as a part of the Data Science Blogathon.

About the Flight Fare Prediction Dataset

  • Airline: So this column will have all the types of airlines like Indigo, Jet Airways, Air India, and many more.
  • Date_of_Journey: This column will let us know about the date on which the passenger’s journey will start.
  • Source: This column holds the name of the place from where the passenger’s journey will start.
  • Destination: This column holds the name of the place to where passengers wanted to travel.
  • Route: Here we can know about that what is the route through which passengers have opted to travel from his/her source to their destination.
  • Arrival_Time: Arrival time is when the passenger will reach his/her destination.
  • Duration: Duration is the whole period that a flight will take to complete its journey from source to destination.
  • Total_Stops: This will let us know in how many places flights will stop there for the flight in the whole journey.
  • Additional_Info: In this column, we will get information about food, kind of food, and other amenities.
  • Price: Price of the flight for a complete journey including all the expenses before onboarding.

By employing machine learning algorithms, particularly regression techniques, we aim to predict flight ticket prices accurately. Leveraging Python for data analysis (regression analysis) and utilizing various machine learning models, including linear regression, will allow us to conduct comprehensive flight price prediction analyses. Additionally, with a focus on the Indian aviation market, we can tailor our predictive models to suit the specific dynamics of this region.

Importing Libraries

import numpy as np
import pandas as pd

import matplotlib.pyplot as plt
import seaborn as sns

from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error as mse
from sklearn.metrics import r2_score
from math import sqrt
from sklearn.linear_model import Ridge
from sklearn.linear_model import Lasso
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import KFold
from sklearn.model_selection import train_test_split
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import RandomizedSearchCV

from prettytable import PrettyTable

Reading the Training Data of our Dataset

Exploratory Data Analysis (EDA)

Now here we will be looking at the kind of columns our dataset has.

train_df.columns

Output:

Index(['Airline', 'Date_of_Journey', 'Source', 'Destination', 'Route',
       'Dep_Time', 'Arrival_Time', 'Duration', 'Total_Stops',
       'Additional_Info', 'Price'],
      dtype='object')

Here we can get more information about our dataset

train_df.info()

Output:

To know more about the dataset

train_df.describe()

Output:

Now while using the IsNull function we will gonna see the number of null values in our dataset

train_df.isnull().head()

Output:

Now while using the IsNull function and sum function we will gonna see the number of null values in our dataset

train_df.isnull().sum()

Output:

Airline            0
Date_of_Journey    0
Source             0
Destination        0
Route              1
Dep_Time           0
Arrival_Time       0
Duration           0
Total_Stops        1
Additional_Info    0
Price              0
dtype: int64

Dropping NAN values

train_df.dropna(inplace = True)

Duplicate values

train_df[train_df.duplicated()].head()

Output:

Here we will be removing those repeated values from the dataset and keeping the in-place attribute to be true so that there will be no changes.

train_df.drop_duplicates(keep='first',inplace=True)
train_df.head()

Output:

train_df.shape

Output:

(10462, 11)

Checking the Additional_info column and having the count of unique types of values.

train_df["Additional_Info"].value_counts()

Output:

No info                         8182
In-flight meal not included     1926
No check-in baggage included     318
1 Long layover                    19
Change airports                    7
Business class                     4
No Info                            3
1 Short layover                    1
2 Long layover                     1
Red-eye flight                     1
Name: Additional_Info, dtype: int64

Checking the different Airlines

train_df["Airline"].unique()

Output:

array(['IndiGo', 'Air India', 'Jet Airways', 'SpiceJet',
       'Multiple carriers', 'GoAir', 'Vistara', 'Air Asia',
       'Vistara Premium economy', 'Jet Airways Business',
       'Multiple carriers Premium economy', 'Trujet'], dtype=object)

Checking the different Airline Routes

train_df["Route"].unique()

Output: See the code.

Now let’s look at our testing dataset

test_df = pd.read_excel("Test_set.xlsx")
test_df.head(10)

Output:

Now here we will be looking at the kind of columns our testing data has.

test_df.columns

Output:

Index(['Airline', 'Date_of_Journey', 'Source', 'Destination', 'Route',
       'Dep_Time', 'Arrival_Time', 'Duration', 'Total_Stops',
       'Additional_Info'],
      dtype='object')

Information about the dataset

test_df.info()

Output:

To know more about the testing dataset

test_df.describe()

Output:

To know more about the testing dataset

Now while using the IsNull function and sum function we will gonna see the number of null values in our testing data

test_df.isnull().sum()

Output:

Airline            0
Date_of_Journey    0
Source             0
Destination        0
Route              0
Dep_Time           0
Arrival_Time       0
Duration           0
Total_Stops        0
Additional_Info    0
dtype: int64

Data Visualization of Flight Price Prediction using Machine Learning

Plotting Price vs Airline plot

sns.catplot(y = "Price", x = "Airline", data = train_df.sort_values("Price", ascending = False), kind="boxen", height = 8, aspect = 3)
plt.show()

Output:

Plotting Price vs Airline plot

Inference: Here with the help of the cat plot we are trying to plot the boxplot between the price of the flight and the airline and we can conclude that Jet Airways has the most outliers in terms of price.

Plotting Violin plot for Price vs Source

sns.catplot(y = "Price", x = "Source", data = train_df.sort_values("Price", ascending = False), kind="violin", height = 4, aspect = 3)
plt.show()

Output:

Plotting Violin plot for Price vs Source

Inference: Now with the help of cat plot only we are plotting a box plot between the price of the flight and the source place i.e. the place from where passengers will travel to the destination and we can see that Banglore is the source location has the most outliers while Chennai has the least.

Plotting Box plot for Price vs Destination

sns.catplot(y = "Price", x = "Destination", data = train_df.sort_values("Price", ascending = False), kind="box", height = 4, aspect = 3)
plt.show()

Output:

Plotting Box plot for Price vs Destination

Inference: Here we are plotting the box plot with the help of a cat plot between the price of the flight and the destination to which the passenger is traveling and figured out that New Delhi has the most outliers and Kolkata has the least.

Feature Engineering

Let’s see our processed data first

train_df.head()

Output:

Processed Data

Here first we are dividing the features and labels and then converting the hours in minutes.

train_df['Duration'] = train_df['Duration'].str.replace("h", '*60').str.replace(' ','+').str.replace('m','*1').apply(eval)
test_df['Duration'] = test_df['Duration'].str.replace("h", '*60').str.replace(' ','+').str.replace('m','*1').apply(eval)

Date_of_Journey: Here we are organizing the format of the date of journey in our dataset for better preprocessing in the model stage.

train_df["Journey_day"] = train_df['Date_of_Journey'].str.split('/').str[0].astype(int)
train_df["Journey_month"] = train_df['Date_of_Journey'].str.split('/').str[1].astype(int)
train_df.drop(["Date_of_Journey"], axis = 1, inplace = True)

Dep_Time: Here we are converting departure time into hours and minutes

train_df["Dep_hour"] = pd.to_datetime(train_df["Dep_Time"]).dt.hour
train_df["Dep_min"] = pd.to_datetime(train_df["Dep_Time"]).dt.minute
train_df.drop(["Dep_Time"], axis = 1, inplace = True)

Arrival_Time: Similarly we are converting the arrival time into hours and minutes.

train_df["Arrival_hour"] = pd.to_datetime(train_df.Arrival_Time).dt.hour
train_df["Arrival_min"] = pd.to_datetime(train_df.Arrival_Time).dt.minute
train_df.drop(["Arrival_Time"], axis = 1, inplace = True)

Now after final preprocessing let’s see our dataset

train_df.head()

Output:

Now after final preprocessing let’s see our dataset

Plotting Bar chart for Months (Duration) vs Number of Flights

plt.figure(figsize = (10, 5))
plt.title('Count of flights month wise')
ax=sns.countplot(x = 'Journey_month', data = train_df)
plt.xlabel('Month')
plt.ylabel('Count of flights')
for p in ax.patches:
    ax.annotate(int(p.get_height()), (p.get_x()+0.25, p.get_height()+1), va='bottom', color= 'black')

Output:

flight price prediction

Inference: Here in the above graph we have plotted the count plot for journey in a month vs several flights and got to see that May has the most number of flights.

Plotting Bar chart for Types of Airline vs Number of Flights

plt.figure(figsize = (20,5))
plt.title('Count of flights with different Airlines')
ax=sns.countplot(x = 'Airline', data =train_df)
plt.xlabel('Airline')
plt.ylabel('Count of flights')
plt.xticks(rotation = 45)
for p in ax.patches:
    ax.annotate(int(p.get_height()), (p.get_x()+0.25, p.get_height()+1), va='bottom', color= 'black')

Output:

flight price prediction using machine learning

Inference: Now from the above graph we can see that between the type of airline and count of flights we can see that Jet Airways has the most flight boarded.

Plotting Ticket Prices vs Airlines

plt.figure(figsize = (15,4))
plt.title('Price VS Airlines')
plt.scatter(train_df['Airline'], train_df['Price'])
plt.xticks
plt.xlabel('Airline')
plt.ylabel('Price of ticket')
plt.xticks(rotation = 90)

Output:

flight price prediction using machine learning

Correlation Between all Features

Plotting Correlation

plt.figure(figsize = (15,15))
sns.heatmap(train_df.corr(), annot = True, cmap = "RdYlGn")
plt.show()

Output:

flight price prediction using machine learning

Dropping the Price column as it is of no use

data = train_df.drop(["Price"], axis=1)

Dealing with Categorical Data and Numerical Data

train_categorical_data = data.select_dtypes(exclude=['int64', 'float','int32'])
train_numerical_data = data.select_dtypes(include=['int64', 'float','int32'])

test_categorical_data = test_df.select_dtypes(exclude=['int64', 'float','int32','int32'])
test_numerical_data  = test_df.select_dtypes(include=['int64', 'float','int32'])
train_categorical_data.head()

Output:

flight price prediction using machine learning

Label Encode and Hot Encode for Categorical Columns

le = LabelEncoder()
train_categorical_data = train_categorical_data.apply(LabelEncoder().fit_transform)
test_categorical_data = test_categorical_data.apply(LabelEncoder().fit_transform)
train_categorical_data.head()

Output:

Concatenating both Categorical Data and Numerical Data

X = pd.concat([train_categorical_data, train_numerical_data], axis=1)
y = train_df['Price']
test_set = pd.concat([test_categorical_data, test_numerical_data], axis=1)
X.head()

Output:

y.head()

Output:

0     3897
1     7662
2    13882
3     6218
4    13302
Name: Price, dtype: int64
# Calculating Mean Absolute Percentage Error
def mean_absolute_percentage_error(y_true, y_pred): 
    y_true, y_pred = np.array(y_true), np.array(y_pred)
    return np.mean(np.abs((y_true - y_pred) / y_true)) * 100

Now we will be splitting out our dataset

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.3, random_state = 42)
print("The size of training input is", X_train.shape)
print("The size of training output is", y_train.shape)
print("The size of testing input is", X_test.shape)
print("The size of testing output is", y_test.shape)

Output:

The size of training input is (7323, 13)
The size of training output is (7323,)
The size of testing input is (3139, 13)
The size of testing output is (3139,)

Model Building

Ridge Regression

# Performing GridSearchCV on Ridge Regression
params = {'alpha' : [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000, 100000]}
ridge_regressor = GridSearchCV(Ridge(), params, cv = 5, scoring = 'neg_mean_absolute_error', n_jobs = -1)
ridge_regressor.fit(X_train, y_train)

Output:

GridSearchCV(cv=5, estimator=Ridge(), n_jobs=-1,
param_grid={‘alpha’: [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000, 100000]},
scoring=’neg_mean_absolute_error’)

# Predicting train and test results
y_train_pred = ridge_regressor.predict(X_train)
y_test_pred = ridge_regressor.predict(X_test)
print("Train Results for Ridge Regressor Model:")
print("Root Mean Squared Error: ", sqrt(mse(y_train.values, y_train_pred)))
print("Mean Absolute % Error: ", round(mean_absolute_percentage_error(y_train.values, y_train_pred)))
print("R-Squared: ", r2_score(y_train.values, y_train_pred))

Output:

Train Results for Ridge Regressor Model:
Root Mean Squared Error: 3558.667750232805
Mean Absolute % Error: 32
R-Squared: 0.4150529285926381

print("Test Results for Ridge Regressor Model:")
print("Root Mean Squared Error: ", sqrt(mse(y_test, y_test_pred)))
print("Mean Absolute % Error: ", round(mean_absolute_percentage_error(y_test, y_test_pred)))
print("R-Squared: ", r2_score(y_test, y_test_pred))

Output:

Test Results for Ridge Regressor Model:
Root Mean Squared Error: 3457.5985597925214
Mean Absolute % Error: 32
R-Squared: 0.42437171409958274

Lasso Regression

# Performing GridSearchCV on Lasso Regression
params = {'alpha' : [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000, 100000]}
lasso_regressor = GridSearchCV(Lasso(), params ,cv = 15,scoring = 'neg_mean_absolute_error', n_jobs = -1)
lasso_regressor.fit(X_train, y_train)

Output:

GridSearchCV(cv=15, estimator=Lasso(), n_jobs=-1,
param_grid={‘alpha’: [0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1000, 10000, 100000]}, scoring=’neg_mean_absolute_error’)

# Predicting train and test results
y_train_pred = lasso_regressor.predict(X_train)
y_test_pred = lasso_regressor.predict(X_test)
print("Train Results for Lasso Regressor Model:")
print("Root Mean Squared Error: ", sqrt(mse(y_train.values, y_train_pred)))
print("Mean Absolute % Error: ", round(mean_absolute_percentage_error(y_train.values, y_train_pred)))
print("R-Squared: ", r2_score(y_train.values, y_train_pred))

Output:

Train Results for Lasso Regressor Model:
Root Mean Squared Error: 3560.853987663486
Mean Absolute % Error: 32
R-Squared: 0.4143339932536655

print("Test Results for Lasso Regressor Model:")
print("Root Mean squared Error: ", sqrt(mse(y_test, y_test_pred)))
print("Mean Absolute % Error: ", round(mean_absolute_percentage_error(y_test, y_test_pred)))
print("R-Squared: ", r2_score(y_test, y_test_pred))

Output:
Test Results for Lasso Regressor Model: Root Mean squared Error: 3459.384927631988 Mean Absolute % Error: 32 R-Squared: 0.4237767638929625

Decision Tree Regression

# Performing GridSearchCV on Decision Tree Regression
depth = list(range(3,30))
param_grid = dict(max_depth = depth)
tree = GridSearchCV(DecisionTreeRegressor(), param_grid, cv = 10)
tree.fit(X_train,y_train)

Output:

GridSearchCV(cv=10, estimator=DecisionTreeRegressor(),
param_grid={‘max_depth’: [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29]})

# Predicting train and test results
y_train_pred = tree.predict(X_train)
y_test_pred = tree.predict(X_test)
print("Train Results for Decision Tree Regressor Model:")
print("Root Mean squared Error: ", sqrt(mse(y_train.values, y_train_pred)))
print("Mean Absolute % Error: ", round(mean_absolute_percentage_error(y_train.values, y_train_pred)))
print("R-Squared: ", r2_score(y_train.values, y_train_pred))

Output:

Train Results for Decision Tree Regressor Model:
Root Mean squared Error: 560.9099093439073
Mean Absolute % Error: 3
R-Squared: 0.9854679156224377

print("Test Results for Decision Tree Regressor Model:")
print("Root Mean Squared Error: ", sqrt(mse(y_test, y_test_pred)))
print("Mean Absolute % Error: ", round(mean_absolute_percentage_error(y_test, y_test_pred)))
print("R-Squared: ", r2_score(y_test, y_test_pred))

Output:

Test Results for Decision Tree Regressor Model:
Root Mean Squared Error: 1871.5387049259973
Mean Absolute % Error: 9
R-Squared: 0.8313483417949448

ridge_score = round(ridge_regressor.score(X_train, y_train) * 100, 2)
ridge_score_test = round(ridge_regressor.score(X_test, y_test) * 100, 2)

lasso_score = round(lasso_regressor.score(X_train, y_train) * 100, 2)
lasso_score_test = round(lasso_regressor.score(X_test, y_test) * 100, 2)

decision_score = round(tree.score(X_train, y_train) * 100, 2)
decision_score_test = round(tree.score(X_test, y_test) * 100, 2)

Comparing all the Models

# Comparing all the models
models = pd.DataFrame({
    'Model': [ 'Ridge Regression', 'Lasso Regression','Decision Tree Regressor'],
    'Score': [ ridge_score, lasso_score, decision_score],
    'Test Score': [ ridge_score_test, lasso_score_test, decision_score_test]})
models.sort_values(by='Test Score', ascending=False)

Output:

Model Score Test Score
Decision Tree Regressor 98.55 83.13
Lasso Regression -252062.50 -248119.29
Ridge Regression -252539.70 -248538.03

# Training = Tr.
# Testing = Te.
x = PrettyTable()
x.field_names = ["Model Name", "Tr. RMSE", "Tr. MA%E", "Tr. R-Squared", "Te. RMSE", "Te. MA%E", "Te. R-Squared",]
x.add_row(['Ridge Regression','3558.67','32','0.42','3457.60','32','0.42'])
x.add_row(['Lasso Regression','3560.85','32','0.41','3459.38','32','0.42'])
x.add_row(['Decision Tree Regressor','853.54','06','0.97','1857.68','10','0.83'])
print(x)

Output:

+————————-+———-+———-+—————+———-+———-+—————+

| Model Name | Tr. RMSE | Tr. MA%E | Tr. R-Squared | Te. RMSE | Te. MA%E | Te. R-Squared |

+————————-+———-+———-+—————+———-+———-+—————+

| Ridge Regression | 3558.67 | 32 | 0.42 | 3457.60 | 32 | 0.42 |

| Lasso Regression | 3560.85 | 32 | 0.41 | 3459.38 | 32 | 0.42 |

| Decision Tree Regressor | 853.54 | 06 | 0.97 | 1857.68 | 10 | 0.83 |

+————————-+———-+———-+—————+———-+———-+—————+

Conclusion

With this, we come to an end of our article – flight price prediction using machine learning. Our regression models have successfully forecasted airline ticket prices with notable accuracy. Through rigorous feature engineering and optimization, particularly in decision tree regression, we’ve gained valuable insights into market dynamics.

As AI continues to evolve, machine learning techniques play a crucial role in accurately predicting airfare prices. Ensemble methods like random forest hold promise for further improving prediction accuracy, ensuring robustness in our models.

In summary, our study demonstrates the effectiveness of machine learning in forecasting airfare prices. Continued advancements in deep learning techniques will likely lead to even more precise predictions, benefiting travelers and industry stakeholders alike.

Here’s the repo link to this article. Hope you liked my article on flight fare prediction using machine learning. If you have any opinions or questions, then comment below.


Frequently Asked Questions

Q1. How does hyperparameter tuning impact the flight fare prediction model?

A. Hyperparameter tuning optimizes the performance of machine learning algorithms by adjusting parameters like alpha in Ridge Regression or max_depth in Decision Tree Regression. It enhances the model’s accuracy and generalization by finding the best parameter values through techniques like GridSearchCV.

Q2. In what way does artificial intelligence improve the accuracy of flight ticket price forecasts?

A. Artificial intelligence, especially machine learning algorithms, analyzes historical flight data to learn patterns and make accurate predictions. By leveraging regression algorithms, AI captures complex relationships between features like airline, departure time, and destination, leading to more precise forecasts.

Q3. What is bagging, and how does it relate to predicting airline ticket prices?

A. Bagging (Bootstrap Aggregating) combines multiple models trained on different subsets of the training data. In predicting ticket prices, it involves training multiple decision tree regressors on different data subsets and averaging their predictions to reduce variance and enhance accuracy.

Q4. What validation methods were used in the flight fare prediction study?

A. Validation techniques such as cross-validation and GridSearchCV were employed. These methods assess model performance by splitting the dataset into subsets for training and testing, and by searching for the best hyperparameters. Metrics like RMSE and R-squared were used to evaluate performance.

Q5. Why is feature selection crucial for improving predictor performance in ticket price forecasting?

A. Feature selection identifies the most relevant features impacting ticket prices, reducing model complexity and improving generalization. By focusing on influential factors and eliminating irrelevant ones, feature selection enhances model accuracy and interpretability, leading to better forecasts.

The media shown in this article is not owned by Analytics Vidhya and is used at the Author’s discretion.

Aman Preet Gulati 19 Feb 2024

Frequently Asked Questions

Lorem ipsum dolor sit amet, consectetur adipiscing elit,

Responses From Readers

Clear

Related Courses

image.name
0 Hrs 19 Lessons
4.84

Big Mart Sales Prediction Using R

Free

image.name
0 Hrs 16 Lessons
4.86

Loan Prediction Practice Problem (Using Python)

Free

  • [tta_listen_btn class="listen"]