Time series analysis looks at data collected over time. It helps find patterns and trends in this data. It uses past data to estimate future values. In this article, I will explain time series analysis.
Key Components
- Trend: The trend shows the overall direction of the data over time. It can be upward, downward, or flat.
- Seasonality: Seasonality refers to patterns that repeat at regular times, like every month or year.
- Cycle: The cycle shows long-term ups and downs that are often linked to economic changes. These cycles can last several years.
- Noise: Noise represents random variations in the data that don’t fit with trends, seasonality, or cycles.
Time Series Modelling
ARIMA Models
ARIMA stands for AutoRegressive Integrated Moving Average. AutoRegressive (AR) uses past data to predict future values. Integrated (I) makes data stable by removing trends and seasonality. Moving Average (MA) corrects predictions based on past errors.
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from statsmodels.tsa.arima.model import ARIMA
# Create a synthetic dataset
np.random.seed(50)
dates = pd.date_range(start='2024-01-01', periods=100, freq='D')
values = np.sin(np.linspace(0, 3 * np.pi, 100)) + np.random.normal(size=100)
data = pd.DataFrame({'Date': dates, 'Value': values})
data.set_index('Date', inplace=True)
# Fit ARIMA model
model = ARIMA(data['Value'], order=(5, 1, 0))
fit = model.fit()
# Forecast
forecast_steps = 10
forecast = fit.forecast(steps=forecast_steps)
forecast_index = pd.date_range(start=data.index[-1] + pd.Timedelta(days=1), periods=forecast_steps)
# Plot the results
plt.figure(figsize=(8, 6))
plt.plot(data.index, data['Value'], label='Historical Data', color='blue')
plt.plot(forecast_index, forecast, label='Forecast', color='red')
plt.title('ARIMA Forecast')
plt.xlabel('Date')
plt.ylabel('Value')
plt.legend()
plt.grid(True)
plt.show()
The output shows a plot with two lines. The blue line is the historical data for 100 days. The red line is the forecast for the next 10 days. This plot shows how the ARIMA model predicts future values using past data.
ETS
ETS stands for Error, Trend, and Seasonality. It breaks data into three parts: error, trend, and seasonality. The error covers random noise. The trend shows long-term changes. Seasonality tracks repeating patterns.
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from statsmodels.tsa.holtwinters import ExponentialSmoothing
# Create a synthetic dataset
np.random.seed(50)
dates = pd.date_range(start='2024-01-01', periods=100, freq='M')
values = np.sin(np.linspace(0, 3 * np.pi, 100)) + np.random.normal(size=100)
data = pd.DataFrame({'Date': dates, 'Value': values})
data.set_index('Date', inplace=True)
# Fit ETS model
model = ExponentialSmoothing(data['Value'], seasonal='add', seasonal_periods=12)
fit = model.fit()
# Forecast
forecast_steps = 10
forecast = fit.forecast(steps=forecast_steps)
forecast_index = pd.date_range(start=data.index[-1] + pd.DateOffset(months=1), periods=forecast_steps, freq='M')
# Plot the results
plt.figure(figsize=(8, 6))
plt.plot(data.index, data['Value'], label='Historical Data', color='blue')
plt.plot(forecast_index, forecast, label='Forecast', color='red')
plt.title('ETS Forecast')
plt.xlabel('Date')
plt.ylabel('Value')
plt.legend()
plt.grid(True)
plt.show()
The output shows a plot with two lines. The blue line represents the historical data for 100 months. The red line shows the forecast for the next 10 months. This plot demonstrates how the ETS model predicts future values based on past data.
STL
STL stands for Seasonal and Trend decomposition using Loess. It breaks time series data into three parts: trend, seasonal, and residual. The trend shows long-term movement. The seasonal part captures repeating patterns. The residual covers random fluctuations.
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from statsmodels.tsa.seasonal import STL
# Create a synthetic dataset
np.random.seed(50)
dates = pd.date_range(start='2024-01-01', periods=100, freq='M')
values = np.sin(np.linspace(0, 3 * np.pi, 100)) + np.random.normal(size=100)
data = pd.DataFrame({'Date': dates, 'Value': values})
data.set_index('Date', inplace=True)
# Decompose the time series
stl = STL(data['Value'], seasonal=13)
result = stl.fit()
# Plot decomposition
plt.figure(figsize=(8, 6))
result.plot()
plt.show()
The output shows a plot with three components. The top part shows the original data and its trend. The middle part shows the seasonal component with repeating patterns. The bottom part shows the residual component with random noise.
Model Evaluation and Validation
Cross-Validation Techniques
Cross-validation in time series involves splitting the data into training and testing sets. It keeps the time order intact. Train the model on past data and test it on future data. This helps you see how well the model works. Common approaches include:
- Rolling-Window Cross-Validation: Train the model on a fixed-size window that moves forward. Each step uses the previous window for training and the next segment for testing.
- Expanding-Window Cross-Validation: The training set gets bigger over time. The test set stays the same size.
Performance Metrics
- Mean Absolute Error (MAE): Tells the average size of mistakes in predictions.
- Mean Squared Error (MSE): Calculates the average of squared mistakes.
- Root Mean Squared Error (RMSE): Takes the square root of the Mean Squared Error.
- Mean Absolute Percentage Error (MAPE): Shows mistakes as a percentage of actual values.
Practical Applications
Financial Forecasting
Time series analysis predicts future stock prices, revenue, or expenses. Studying past financial data helps you make better investment choices. It also helps with managing risks and planning budgets. This method shows trends and helps businesses adjust plans based on future forecasts.
Demand Forecasting
Businesses use time series analysis to predict future product demand. This helps them manage inventory and supply chains better. By looking at past sales, they can forecast demand more accurately. This prevents having too much or too little stock. It also helps improve customer satisfaction by avoiding stockouts and overstocking.
Climate and Weather Prediction
Time series analysis helps forecast weather and study climate trends. Meteorologists use it to give daily and weekly weather updates. This helps people and organizations plan and prepare for weather. Climate scientists use it to track changes in temperature and rainfall. This information guides climate policy and environmental plans.
Healthcare
Time series analysis is important in healthcare. It helps predict disease outbreaks and manage hospital admissions. By looking at past health data, officials can forecast disease spread and plan responses. Hospitals use forecasts to predict patient admissions and adjust staffing. This also helps improve patient care and estimate recovery times.
Energy Sector
Time series analysis is used to predict energy use and plan for renewable energy. It helps ensure energy production matches demand, avoiding shortages or surpluses. For wind and solar energy, forecasts help balance the grid. It also helps schedule maintenance to keep the energy system reliable and efficient.
Wrapping Up
Time series analysis helps us see how data changes over time. It shows trends, patterns, and seasonal changes. This makes it easier to predict future values. It is useful in finance, business, weather, healthcare, and energy. Using time series analysis helps you make better decisions and plan effectively.
Image generated with Midjourney.
