Air Temperature Forecasting

import kagglehub

# Download latest version
path = kagglehub.dataset_download("melaniaberbatova/time-series-data")

print("Path to dataset files:", path)

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import TimeSeriesSplit
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import make_scorer
import os

os.listdir(path)

df_train = pd.read_csv(os.path.join(path,"daily_climate_train.csv"))
df_train.head(3)

df_train.tail(3)

# General statistics of the data
df_train.describe()

# Correlation matrix
correlation_mat = df_train.select_dtypes('number').corr()
fig, ax = plt.subplots()
sns.heatmap(correlation_mat, annot = True, ax=ax)
plt.show()

# Distributions and joint distributions of the data
sns.pairplot(df_train)

df_test = pd.read_csv(os.path.join(path,"daily_climate_test.csv"))
df_test.head()

# Start your solution here

# Feature construction
# Add additional features to the data here

# Construction of the training set
# Do not change this code!

# Removing the date column, since RandomForestRegressor does not support the format
df_train.drop(columns='date', inplace=True)

X_train = df_train.drop('meantemp', axis=1)
y_train = df_train['meantemp']

# Splitting the data for cross-validation
tscv = TimeSeriesSplit(n_splits=5)

# Creating a grid for the hyperparameter search
param_grid = {
    'n_estimators': [50, 100, 200],
    # change the values and/or add more hyperparameters here
}

# Initialising the scoring function and the model
# Do not change this code!
scorer = make_scorer(mean_squared_error, greater_is_better=False)

rf = RandomForestRegressor(random_state=42, n_jobs=-1)

# Hyperparameter tuning
grid_search = GridSearchCV(estimator=rf,
                          param_grid=param_grid,
                          cv=tscv,
                          scoring=scorer,
                          n_jobs=-1,
                          verbose=1)
grid_search.fit(X_train, y_train)

# Model selection
# Do not change this code!
model = grid_search.best_estimator_
print(f"Best parameters: {grid_search.best_params_}")

# Add the same features to the test set as well
# The number of columns in X_test must be the same as in X_train
X_test = df_test.copy()

# Prediction on the test set
# Do not change this code!
X_test = df_test.copy()
X_test.drop(columns='date', inplace=True)
predictions = model.predict(X_test)

# Saving the predictions to the file submission_time_series.csv
# Do not change this code!
submission_df = pd.DataFrame({'predictions': predictions})
submission_df.to_csv('submission_time_series.csv', index=False)