The purpose is discovering knowledge from timeseries data which is stored pairs of measurement time and value such as accumlated
data from sensors. For example, we develop methods for prediction of future behavior of stock price
data, forecasting of condition of the patient from body templature and blood element.
In the case of prediction of generation power of photovoltaic power generation, rules are extracted by combination of value
of solar radiation, templature and humidity. The graph of solar radiation is
Extracted rules are represented by IF-THEN style for easy to understand.
Introduce several softwares.
Here is visualize retrieveral system for timeseries database．
It retrieves similar behaviors into database, and extracts them into csv format file.
In the query dialog, a user is drawable on a graph of a behavior pattern by using any pointing device. Pattern matching is calculated with Dynamic Time Warping．
In the result dialog, start time of matched period is listed. Listed items are able to choose and show. And also, it can extract just matched periods into CSV format.
Here is a tool for analysis of timeseries data with combination of multiple shape. The application is able to carries out high flexible retrieval by combination of shape drawing area and original query language. And also obtained data is throwable into machine language engine of Weka.
For initialize and making database, CSV format is importable.
Here is query dialog.
Result dialog shows retrieved time period. The right side shows retrieved time period. The left side bar shows the amount of the database and an indicator for currently showing time period. Status bar shows the number of retrieved periods.
Here is an example of the extracted rules by using desicion tree learning.