Scribd
Upload
8 views15 pages

04 Time Series Forecasting Framework

The document outlines a forecasting framework that includes defining what to forecast, determining the forecasting horizon, and identifying available data. It emphasizes the importance of selecting specific metrics, such as sales or energy demand, and the time frame for predictions. Additionally, it discusses the extraction of features from historical data to inform future forecasts.

Uploaded by

sharma94.chirag
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
8 views15 pages

04 Time Series Forecasting Framework

The document outlines a forecasting framework that includes defining what to forecast, determining the forecasting horizon, and identifying available data. It emphasizes the importance of selecting specific metrics, such as sales or energy demand, and the time frame for predictions. Additionally, it discusses the extraction of features from historical data to inform future forecasts.

Uploaded by

sharma94.chirag
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 15

Forecasting

Framework
Setting up the
approach

1
Forecasting framework

Define:

• What do I want to forecast?

• When do I want the forecast?

• What data do I have available?

2
What to forecast

To start, we need to decide what we want to forecast.

For example:

• Sales per country/product vs total sales?


• Energy demand per household vs per city?
• Daily temperature or hourly temperature?

3
What to forecast

Sales Sales

Time UK Spain Australia Belgium Time Total


30/03/20 200 100 330 120 30/03/20 450
31/03/20 220 120 300 135 31/03/20 775
01/04/20 230 150 335 133 01/04/20 848
02/04/20 235 175 340 200 02/04/20 950

We may need to group / aggregate our time series.

4
The forecasting horizon
How far in advance we want our predictions / forecast to be?

Forecast point:
• Forecast weekly sales for next week?
• Daily energy demand for tomorrow?
• Hourly stock price next hour?

Forecasting window:
• Forecast weekly sales for next 3 weeks?
• Forecast daily energy demand over the next 7 days?
• Hourly stock price in the next 6 hours?

5
The available data

What data do we have up to (but not including) the time of forecast?

For example:

• Sales revenue up to last week.


• Energy demand up to yesterday.
• Stock prices up to now.

6
Forecasting Framework

Available Forecast
Data point

Time

In the simplest case,


• We have data up to a certain point in time.
• We want to forecast 1 single point ahead.
• E.g., forecast sales next week with weekly sales data up to last week.

7
Forecasting Framework

Available Forecast
Data point

Time

In the simplest case,


• We have data up to a certain point in time.
• We want to forecast 1 single point ahead.
• E.g., forecast sales next week with weekly sales data up to last week.

8
Forecasting Framework

Available Forecasting
Data window

Time

• We have data up to a certain point in time.


• We want to forecast various points ahead.
• E.g., forecast weekly sales in the next 3 weeks, with weekly sales data up to last week.

9
Forecasting Framework

Available Forecasting
Data window

Time

• We have data up to a certain point in time.


• We want to forecast various points ahead.
• E.g., forecast weekly sales in the next 3 weeks, with weekly sales data up to last week.

10
Forecasting Framework

Available Forecast
Data point

Time

t-n … t-2 t-1


time t

• We want to predict a value at time t.


• We can use previous data, that is t-1, t-2, etc.

11
Feature derivation window

Time Sales

30/03/20 200 t-n


31/03/20 220 … Feature
derivation
01/04/20 230 t-2 window
02/04/20
Time 235
Sales t-1
03/04/20 ? t

We extract features only from the feature derivation window.

Forcasting models should be trained on available data.

12
Feature derivation window

Time Sales

30/03/20 200 t-n


31/03/20 220

01/04/20 230 … Feature


derivation
02/04/20
Time 235
Sales t-2 window
03/04/20
Time 220
Sales t-1
03/04/20 ? t

We extract features only from the feature derivation window.

The feature derivation window changes for each time point.

13
Feature derivation window
Predictors

Time Var 1 Var 2 Var 2 Var 4 Var 5 Var 5 Sales

30/03/20 3 30 15 50 0.2 0 200 t-n


31/03/20 3 31 16 10 0.2 0 220 … Feature
01/04/20 4 1 17 0 0.19 1 230
derivation
t-2 window
02/04/20
Time 4 1
Var 2 2
Var 19 2
Var 54
Var 0.175
Var 0 5
Var 235
Sales t-1
03/04/20 4 3 ? ? ? ? ? t

Features that we know  date (independent of derivation window

Future unknown features  can only extract from the feature derivation window.

14
Define what we want to forecast.

Define the forecasting horizon

Summary
Define available data

Future unknown features can be


created only from data available up to
the forecasting point.

15

You might also like