main.py

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn import preprocessing, metrics
import seaborn as sb
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
from sklearn.preprocessing import PolynomialFeatures
from sklearn.linear_model import LinearRegression, Lasso

""" The following code estimates flight delay. It includes preprocessing data, its visualization, detecting and removing
 outliers. Then, it uses 3 machine learning models to predict the flight delay. They are: linear regression, polynomial 
    regression and lasso regularization. After applying these models, the code checks its metrics. """

# accessing dataset of flight_delay.csv
dataset = pd.read_csv('flight_delay.csv')
print(f'{dataset.head().to_string()}\n')


# encoding data to obtain categorical features
label_enc = preprocessing.LabelEncoder()
dataset['Depature Airport'] = label_enc.fit_transform(dataset['Depature Airport'])
dataset['Destination Airport'] = label_enc.fit_transform(dataset['Destination Airport'])
print(f'{dataset.head().to_string()}\n')


# checking for missing values
dataset.isnull()
dataset.isnull().sum()

# converting string of the following columns into datetime
dataset['Scheduled depature time'] = pd.to_datetime(dataset['Scheduled depature time'])
dataset['Scheduled arrival time'] = pd.to_datetime(dataset['Scheduled arrival time'])

# adding a new feature, i.e. duration
duration = dataset['Scheduled arrival time'] - dataset['Scheduled depature time']
flight_dur_sec = duration.dt.total_seconds()
flight_dur = flight_dur_sec/60
dataset['Flight duration'] = flight_dur
print(f'{dataset.head().to_string()}\n')


# splitting data into train and test data
train_data = dataset.drop(dataset[pd.DatetimeIndex(dataset['Scheduled depature time']).year == 2018].index)
test_data = dataset.drop(dataset[pd.DatetimeIndex(dataset['Scheduled depature time']).year < 2018].index)

# plotting correlation matrix to see the delay dependence of different features
corr_mat = train_data.corr()
sb.heatmap(corr_mat, annot = True)
plt.show()


# visualising data
plt.plot(dataset['Flight duration'], dataset['Delay'], 'bo')
plt.title('Flight duration vs Delay')
plt.xlabel('Flight duration (min)')
plt.ylabel('Delay (min)')
plt.show()


# define the outlier
sb.boxplot(data = dataset['Delay'], x = dataset['Flight duration'])
plt.title('Boxplot for outliers detection')


# outliers removing
train_data = train_data.drop(train_data[(train_data['Flight duration'] > 900)].index)


# obtaining X_train, y_train, x_test, y_test
X_train = train_data.drop(['Scheduled depature time', 'Scheduled arrival time','Delay'], axis=1)
y_train = train_data['Delay']
x_test = test_data.drop(['Scheduled depature time', 'Scheduled arrival time','Delay'], axis=1)
y_test = test_data['Delay']


# Linear Regression
regressor = LinearRegression()
regressor.fit(X_train, y_train)

# checking train data
y_train_pred = regressor.predict(X_train)

# checking test data
y_test_pred = regressor.predict(x_test)


print('Metrics for Linear Regression')

# checking train data
print('Testing train data')
print('Mean Absolute Error:', metrics.mean_absolute_error(y_train, y_train_pred))
print('Mean Squared Error:', metrics.mean_squared_error(y_train, y_train_pred))
print('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_train, y_train_pred)))
print('R score', metrics.r2_score(y_train, y_train_pred))


# checking test data
print('Testing test data')
print('Mean Absolute Error:', metrics.mean_absolute_error(y_test, y_test_pred))
print('Mean Squared Error:', metrics.mean_squared_error(y_test, y_test_pred))
print('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_test, y_test_pred)))
print('R score', metrics.r2_score(y_test, y_test_pred))


# Polynomial Regression
poly_reg = PolynomialFeatures(degree=3)
x_train_poly = poly_reg.fit_transform(X_train)
poly_reg_model = LinearRegression()
poly_reg_model.fit(x_train_poly,y_train)

# checking train data
y_train_pred = poly_reg_model.predict(x_train_poly)

# checking test data
y_test_pred = poly_reg_model.predict(poly_reg.fit_transform(x_test))

print('Metrics for polynomial regression')

# checking train data
print('Testing train data')
print('Mean Absolute Error:', metrics.mean_absolute_error(y_train, y_train_pred))
print('Mean Squared Error:', metrics.mean_squared_error(y_train, y_train_pred))
print('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_train, y_train_pred)))
print('R score', metrics.r2_score(y_train, y_train_pred))


# checking test data
print('Testing test data')
print('Mean Absolute Error:', metrics.mean_absolute_error(y_test, y_test_pred))
print('Mean Squared Error:', metrics.mean_squared_error(y_test, y_test_pred))
print('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_test, y_test_pred)))
print('R score', metrics.r2_score(y_test, y_test_pred))



# Lasso

X_train, x_val, y_train, y_val = train_test_split(X_train, y_train, test_size=1/8, random_state=123)

alphas = [2.2, 2, 1.5, 1.3, 1.2, 1.1, 1, 0.3, 0.1]
losses = []

# defining the best alpha for Lasso
for alpha in alphas:
    lasso = Lasso(alpha=alpha)
    lasso.fit(X_train, y_train)
    y_pred = lasso.predict(x_val)
    mse = mean_squared_error(y_val, y_pred)
    losses.append(mse)
plt.plot(alphas, losses)
plt.title("Lasso alpha value selection")
plt.xlabel("alpha")
plt.ylabel("Mean squared error")
plt.show()
best_alpha = alphas[np.argmin(losses)]
print("Best value of alpha:", best_alpha)

lasso = Lasso(best_alpha)
lasso.fit(X_train, y_train)

# checking train data
y_train_pred = lasso.predict(X_train)

# checking test data
y_test_pred = lasso.predict(x_test)

print('Metrics for Lasso')

# checking train data
print('Testing train data')
print('Mean Absolute Error:', metrics.mean_absolute_error(y_train, y_train_pred))
print('Mean Squared Error:', metrics.mean_squared_error(y_train, y_train_pred))
print('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_train, y_train_pred)))
print('R score', metrics.r2_score(y_train, y_train_pred))


# checking test data
print('Testing test data')
print('Mean Absolute Error:', metrics.mean_absolute_error(y_test, y_test_pred))
print('Mean Squared Error:', metrics.mean_squared_error(y_test, y_test_pred))
print('Root Mean Squared Error:', np.sqrt(metrics.mean_squared_error(y_test, y_test_pred)))
print('R score', metrics.r2_score(y_test, y_test_pred))