Time Series Forecasting
Predicting future values of a sequence measured over time, such as sales or electricity demand, from its own past patterns: trend, seasonality and noise.
Also ARIMA · exponential smoothing · Holt-Winters · seasonality · autoregression · forecasting
ARIMA: Box and Jenkins, 1970
If an ice cream shop sold more every summer and a bit more each year, you can guess next July's sales from that pattern before anyone tells you.
Where it fits
+All relationships as a list
Is a kind of / part of
Siblings
Uses / built on
How it works
A time series is a sequence of measurements in time order: daily sales, hourly temperature, monthly passengers. Forecasting asks: what comes next?
Most series mix three ingredients:
- Trend: a slow rise or fall.
- Seasonality: a repeating cycle (weekly, yearly).
- Noise: the unpredictable rest.
Classical models
- Exponential smoothing forecasts with a weighted average of past values, where recent values count more. The Holt-Winters version adds trend and seasonal terms.
- ARIMA combines autoregression (predicting from the last few values, like a small linear regression), integration (differencing the series to remove trends), and a moving average of past forecast errors.
Modern approaches
Gradient boosting on “lag” features (last week’s sales, same day last year) is very strong in practice. RNNs, LSTMs and Transformers handle many related series at once. Pretrained “foundation” forecasting models are also emerging.
The one rule that never changes: test on the future. Shuffle the data randomly and your model will peek at tomorrow to predict yesterday, a classic form of leakage (see train/test split).
Everyday use cases
- Case 01
Retail demand planning
Forecasting how many umbrellas each store needs next week from past sales and the season.
- Case 02
Electricity grids
Predicting tomorrow's hourly power demand so operators can schedule power plants.
Strengths, weaknesses and when to use it
Strengths
- Classical models are fast, interpretable and strong baselines on short univariate series.
Weaknesses
- Cannot foresee breaks in the pattern (a pandemic, a new competitor).
- Easy to evaluate wrongly by shuffling time.
When to use itAny prediction where the order of observations matters. Start with seasonal-naive and exponential smoothing baselines.
Compare with
Check yourself
- Q01
What does the "I" (integrated) part of ARIMA handle?
Show answer
A: Differencing to remove trends
- Q02
How should you split a time series for testing?
Show answer
B: Train on earlier data and test on later data