Advanced

Time Series Analysis

Time series data has a time dimension that introduces unique challenges and opportunities. Learn to decompose trends, detect seasonality, test for stationarity, and build forecasting models.

Time Series Components

Every time series can be decomposed into fundamental components:

📈

Trend

Long-term direction — is the series generally increasing, decreasing, or flat over time?

🔃

Seasonality

Regular, repeating patterns at fixed intervals — monthly sales spikes, daily website traffic patterns.

🔀

Residuals (Noise)

Random fluctuations left after removing trend and seasonality. Ideally, residuals look like white noise.

Python 
import pandas as pd
from statsmodels.tsa.seasonal import seasonal_decompose
import matplotlib.pyplot as plt

# Load time series data
df = pd.read_csv('monthly_sales.csv', parse_dates=['date'],
                  index_col='date')

# Decompose the time series
decomposition = seasonal_decompose(df['sales'], model='additive',
                                    period=12)

# Plot components
fig = decomposition.plot()
fig.set_size_inches(12, 8)
plt.tight_layout()
plt.show()

Stationarity

A stationary time series has constant statistical properties (mean, variance, autocorrelation) over time. Most forecasting models require stationarity.

Python 
from statsmodels.tsa.stattools import adfuller

# Augmented Dickey-Fuller test for stationarity
result = adfuller(df['sales'])
print(f"ADF Statistic: {result[0]:.4f}")
print(f"p-value: {result[1]:.4f}")

if result[1] < 0.05:
    print("Series is stationary")
else:
    print("Series is NOT stationary - apply differencing")

# Make stationary by differencing
df['sales_diff'] = df['sales'].diff().dropna()
💡
Making data stationary: Common techniques include differencing (subtracting the previous value), log transformation (stabilizes variance), and seasonal differencing (subtracting the value from one season ago).

Moving Averages

Moving averages smooth out short-term fluctuations to reveal underlying trends. They are also used as baselines for forecasting.

Python 
# Simple Moving Average
df['SMA_7'] = df['sales'].rolling(window=7).mean()
df['SMA_30'] = df['sales'].rolling(window=30).mean()

# Exponential Moving Average (more weight on recent values)
df['EMA_7'] = df['sales'].ewm(span=7).mean()

# Plot
plt.figure(figsize=(12, 5))
plt.plot(df['sales'], alpha=0.5, label='Actual')
plt.plot(df['SMA_30'], label='30-day SMA')
plt.plot(df['EMA_7'], label='7-day EMA')
plt.legend()
plt.title('Sales with Moving Averages')
plt.show()

ARIMA Models

ARIMA (AutoRegressive Integrated Moving Average) is the most widely used time series forecasting model. It combines three components:

Component Parameter What It Does
AR (AutoRegressive) p Uses past values to predict future values
I (Integrated) d Number of differences needed for stationarity
MA (Moving Average) q Uses past forecast errors to improve predictions
Python 
from statsmodels.tsa.arima.model import ARIMA
import warnings
warnings.filterwarnings('ignore')

# Fit ARIMA model (p=1, d=1, q=1)
model = ARIMA(df['sales'], order=(1, 1, 1))
results = model.fit()
print(results.summary())

# Forecast next 30 days
forecast = results.forecast(steps=30)

# Plot actual vs forecast
plt.figure(figsize=(12, 5))
plt.plot(df['sales'], label='Actual')
plt.plot(forecast, color='red', label='Forecast')
plt.legend()
plt.title('Sales Forecast')
plt.show()

# Auto-select best ARIMA parameters
from pmdarima import auto_arima
auto_model = auto_arima(df['sales'], seasonal=True, m=12,
                        suppress_warnings=True)
print(auto_model.summary())

Exponential Smoothing

Exponential smoothing methods weight recent observations more heavily than older ones. The Holt-Winters method handles trend and seasonality together.

Python 
from statsmodels.tsa.holtwinters import ExponentialSmoothing

# Holt-Winters (Triple Exponential Smoothing)
model = ExponentialSmoothing(
    df['sales'],
    trend='add',          # Additive trend
    seasonal='add',       # Additive seasonality
    seasonal_periods=12  # Monthly data with yearly cycle
)
results = model.fit()

# Forecast
forecast = results.forecast(12)  # Next 12 months

plt.figure(figsize=(12, 5))
plt.plot(df['sales'], label='Actual')
plt.plot(results.fittedvalues, label='Fitted', alpha=0.7)
plt.plot(forecast, color='red', label='Forecast')
plt.legend()
plt.title('Holt-Winters Forecast')
plt.show()

Real-World Applications

Domain Application Common Models
Finance Stock prices, trading volume, risk forecasting GARCH, ARIMA, LSTM
Weather Temperature, rainfall, wind speed prediction SARIMA, state-space models
Retail Demand forecasting, inventory planning Holt-Winters, Prophet, XGBoost
Energy Power consumption, renewable energy output ARIMA, neural networks
Healthcare Disease outbreaks, patient admissions SIR models, SARIMA
✅
Practical tip: For business forecasting, Facebook's Prophet library is excellent — it handles holidays, missing data, and changepoints automatically. Install with pip install prophet.