The ARIMA Model
The ARIMA Model
future values in a time series. It uses past data to make forecasts, such as predicting next month's
sales, temperatures, or stock prices.
1. AR (AutoRegressive) – p
2. I (Integrated) – d
o Removes trends by differencing (subtracting values) to make the data stable over
time.
3. MA (Moving Average) – q
o Example: If the last prediction was off by 2 degrees, the model adjusts for that.
2. Choose p, d, q Values:
o Use plots to help decide how many past values (p) and errors (q) to use, and how
many times to difference (d).
5. Make Forecasts:
2. Adjust for any trends like rising or falling temperatures over time (I part).
The ARIMA model combines these three ideas to give you a better forecast.
Widely Used: Works well for many types of time series data.
When There’s Seasonality: For repeating patterns (e.g., monthly sales), use SARIMA
(Seasonal ARIMA).
When Data is Complicated: For more complex data, you might need advanced methods like
machine learning.
3. Remainder: The random noise left after removing the trend and seasonality.
Steps in STL:
You can extract features like height, average energy, and other statistics from the generated ARIMA
or STL models to analyze and improve predictions.
1. Height:
2. Average Energy:
o The squared sum of the signal values divided by the length, indicating the strength or
intensity of the signal.
3. Trend Slope:
4. Residual Variance:
1. Decompose the Time Series with STL to separate trend, seasonality, and noise.
o Seasonal height/amplitude
o Average energy
o Trend slope
4. Evaluate Forecast Accuracy using metrics like MAE, RMSE, and MAPE.