1. We make ICT strategies work
Prof. Dr.-Ing. Thomas Bauschert , Dr. Mathias Schweigel, Oleksandr Kryvoshapka
Technische Universität Chemnitz, Detecon International GmbH
Feb 2016
A Framework for Telecommunication
Traffic Demand Forecasting
Short- and medium-term forecasts are required for activities that range from operations management to budgeting and selecting new research and development projects.
Long-term forecasts affect issues such as strategic planning.
Short- and medium-term forecasting is typically based on identifying, modeling, and extrapolating the patterns found in historical data.
The task of timing forecasts is to determine the time when an event will happen.
Frequency forecasts are aiming to determine quantities of events that will occur at certain period.
The continuance of an event is the reason for duration forecasts.
Monthly sales for the souvenir shop at a beach resort town in Queensland, Australia [a-little-book-of-r-for-time-series.readthedocs.org]
Periodicity is IMPORTANT!!! Trend
The horizontal direction of a smoothed time-series. Trends can be long-term pattern or dynamic in relatively short-term duration. Trend reflects the underlying growth or decline in the value of the variable. This variation pattern is present at least over several successive periods. Perception of the trend depends on the length of the observed series. If a time series does not show an increasing or decreasing pattern then such time-series called “stationary”.Seasonal variations
Patterns of change in a time-series within a period of no more than a year. These patterns tend to repeat themselves. It refers to short-term, relatively frequent variations, which are identified by the differences between the actual results and the trend line.
In real life, this pattern can repeat hourly, daily, weekly, monthly, yearly, etc. Seasonal variations is always has a known period sometimes called periodic variations. Generally related to factors such as weather, holidays and vacations and so on.
Cyclical variations
The variations of a time-series over periods longer than one year. They are not having a fixed period and often related to the current economic conditions. As a rule, the length of cycles is longer than the length of a season, and the magnitude of cycles usually much higher than the magnitude of seasonal patterns.
Usually cyclical variations are not present in the typical time-series.
Irregular variations
Unpredictable component of every time-series that makes it a random variable. Irregular variations in the data caused by unusual circumstances. In general, the duration if such variations is short.
There are two types of irregular variations can be specified: episodic and residual. Episodic fluctuations can be identified by nature of emergence. The residual fluctuations (chance fluctuations) cannot be identified. Of course, neither episodic nor residual variation can be projected into the future.
The advantages of Periodic DES and Periodic LinReg over normal DES and Linear Regression methods are:
The individual simplicity if original methods is kept.
The new periodic algorithms are able to forecast seasonal time series with (local or global) trend.
The disadvantages are next:
Overall complexity of the method is dependent from the periodicity L of the input data.
More input periods is required to produce adequate forecast.
Operators collect data usually on hourly or sub-hourly basis.To save storage space – aggregation to the higher level.
To forecast next hour can be interesting but not useful.Most used cases in short term forecasting.Forecast MAX to know if we need to increase the capacity of the cell.
Due to the lack of time, only some of the tests will be explained in the details
The main purpose of this test is to find out the impact of different amount of input data on the accuracy of implemented forecasting methods. The preliminary consideration is that more input data used – more accurate (in terms of MSEforecast) will be the forecasted values.
Accuracy measure problem: no distinguish between above or below certain value.