Time Series Forecasting

Business Questions to Answer

The dataset contains hourly energy consumption readings from GridSmart's smart meter network, spanning from January 2022 to January 2025. It includes weather data and temporal features essential for time series analysis.

Dataset Schema

Complete the following steps to build your time series forecasting system. Each step builds upon the previous one, culminating in a comprehensive forecasting solution.

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from statsmodels.tsa.seasonal import seasonal_decompose
from prophet import Prophet

# Load the data
df = pd.read_csv('data/energy_consumption.csv', parse_dates=['date'])
df.set_index('date', inplace=True)
print(f"Dataset shape: {df.shape}")
df.head()

from statsmodels.tsa.stattools import adfuller, kpss

# ADF Test
adf_result = adfuller(daily_data.dropna())
print(f"ADF Statistic: {adf_result[0]:.4f}")
print(f"p-value: {adf_result[1]:.4f}")

Decomposition separates a time series into its fundamental components: trend, seasonality, and residuals. This helps understand underlying patterns and improves forecasting accuracy.

from statsmodels.tsa.seasonal import seasonal_decompose

# Aggregate to daily for decomposition
daily_data = df['consumption_kwh'].resample('D').sum()

# Additive decomposition
result_add = seasonal_decompose(
    daily_data, 
    model='additive', 
    period=7  # Weekly seasonality
)
result_add.plot()
plt.tight_layout()
plt.show()

# Multiplicative decomposition
result_mult = seasonal_decompose(
    daily_data, 
    model='multiplicative', 
    period=7
)
result_mult.plot()
plt.tight_layout()
plt.show()

# Compare residual variance
add_var = result_add.resid.var()
mult_var = result_mult.resid.var()
print(f"Additive Var: {add_var:.2f}")
print(f"Multiplicative Var: {mult_var:.2f}")

Implement both ARIMA and Prophet models to compare their forecasting performance on energy consumption data. Each has strengths for different forecasting scenarios.

ARIMA / SARIMA Model

from statsmodels.tsa.statespace.sarimax import SARIMAX
from statsmodels.graphics.tsaplots import plot_acf, plot_pacf

# Plot ACF and PACF to determine p and q
fig, axes = plt.subplots(1, 2, figsize=(12, 4))
plot_acf(daily_data.dropna(), lags=30, ax=axes[0])
plot_pacf(daily_data.dropna(), lags=30, ax=axes[1])
plt.tight_layout()
plt.show()

# Fit SARIMA model with seasonal component
model = SARIMAX(
    train_data,
    order=(1, 1, 1),
    seasonal_order=(1, 1, 1, 7),  # Weekly seasonality
)
sarima_result = model.fit(disp=False)

# Model diagnostics
sarima_result.plot_diagnostics(figsize=(12, 8))
plt.show()

# Forecast
forecast = sarima_result.forecast(steps=30)
conf_int = sarima_result.get_forecast(steps=30).conf_int()

Facebook Prophet Model

Prophet is designed for business forecasting with automatic handling of seasonality, holidays, and trend changes. It is particularly robust to missing data and outliers.

from prophet import Prophet

# Prepare data for Prophet (requires 'ds' and 'y' columns)
prophet_df = daily_data.reset_index()
prophet_df.columns = ['ds', 'y']

# Create and configure Prophet model
model = Prophet(
    weekly_seasonality=True,
    yearly_seasonality=True,
    changepoint_prior_scale=0.05
)

# Add Indian holidays
model.add_country_holidays(country_name='IN')
model.fit(prophet_df)

# Create future dataframe and forecast
future = model.make_future_dataframe(periods=30)
forecast = model.predict(future)

# Plot forecast and components
model.plot(forecast)
model.plot_components(forecast)
plt.show()

Model Comparison

from sklearn.metrics import mean_absolute_error, mean_squared_error
import numpy as np

def calculate_mape(actual, predicted):
    """Calculate Mean Absolute Percentage Error"""
    return np.mean(np.abs((actual - predicted) / actual)) * 100

def evaluate_model(actual, predicted, model_name):
    """Evaluate forecasting model performance"""
    mae = mean_absolute_error(actual, predicted)
    rmse = np.sqrt(mean_squared_error(actual, predicted))
    mape = calculate_mape(actual, predicted)
    return {'Model': model_name, 'MAE': mae, 'RMSE': rmse, 'MAPE': mape}

# Compare models
comparison_df = pd.DataFrame([
    evaluate_model(test_data, sarima_forecast, 'SARIMA'),
    evaluate_model(test_data, prophet_forecast, 'Prophet')
])
print(comparison_df)

Create the following visualizations to demonstrate your time series analysis and forecasting capabilities. Each visualization should be properly labeled.

Create a public GitHub repository with the exact name shown below:

Required Project Structure

time-series-forecasting/
├── data/
│   └── energy_consumption.csv  # The dataset
├── notebooks/
│   └── time_series_analysis.ipynb  # Your main analysis notebook
├── reports/
│   └── (visualizations)  # Saved plots
├── requirements.txt  # Python dependencies
└── README.md  # Project documentation

requirements.txt

pandas>=2.0.0
numpy>=1.24.0
matplotlib>=3.7.0
seaborn>=0.12.0
statsmodels>=0.14.0
prophet>=1.1.0
jupyter>=1.0.0

Your project will be graded on the following criteria. Total: 600 points.