Unser Projekt:
Alzheimer-Erkrankung: Früherkennung & Entrainment

A simple time series analysis

Toki, the friendly turtle from Tokeya, wants to play a game with you. You simply click on Toki's picture using the left or the right mouse button and Toki will try to guess which button you will push the next time. Alternatively, you can also click one of the two buttons below the picture. Just give it a try and see what happens.


How does our turtle Toki do that?

After each of your clicks a simple time series analysis is performed, i.e., your latest clicks are checked for regularities, of which you are not aware. If, for example, you always click left, right, left, right and so on, Toki recognizes that and will correctly predict your next click. A very simple neural net is trained in the background after each of your clicks to predict your next click. What is interesting here is that to you your clicks may appear to be random, but they often follow certain patterns without you realizing it. Neural nets can find such hidden patterns in time series and allow a prediction of the next event. Neural nets are just one possible method among many for a time series analysis.

A detailed explanation is available on an extra page.

Why is that important?

Simply put, a trustworthy forecast offers substantial competitive advantages and leads to higher profitability. Just as the temporal sequence of your mouse clicks is not completely random, many processes in the economy, nature, or specifically in the industrial condition monitoring follow certain patterns. These include, e.g., the power consumption of a city, the air temperature in a building, or the price of a certain security. In principle, all data from sensors or data bases can be tested for regularities by a time series analysis. Even patterns which cannot be identified with the naked eye or with the help of one's own experience, can be identified by means of a thorough analysis of the respective time series. This allows for a more accurate prediction of future developments.

The methods employed are able to learn from past data, can adapt to it, and can thereby produce constantly improving predictions without any intervention from outside. These predictions can help to answer questions regarding cost optimization (When do we need to replace a component? When should we expect a failure to occur?), customer relationship management (When should customers who bought certain products in the past be offered which services in which way?), or fundamental questions concerning the optimization of warehousing (How large will the demand for a certain product be next month?).

A time series analysis not only provides a simple description of the past, but possibly also reveals an explanation which makes it feasible to compute reliable forecasts and to understand how single events affect the future.