Time Series Data Analysis
Data Mining Research for Knowledge Discovery from Sensor Data
Research Overview
Our research aims to discover knowledge from numerical data recorded over time, such as sensor data.
For example, we work on the following prediction problems:
- Predicting stock price movements from price fluctuation patterns
- Predicting patient conditions from body temperature and blood component data
- Weather prediction from temperature and humidity data
Research Examples
We introduce our specific research content using solar panel power generation prediction as an example.
Solar Power Generation Prediction System
We research rule extraction methods for predicting solar panel power generation from solar irradiance, temperature, and humidity data.
The extracted rules are expressed in human-readable IF-THEN format and can be utilized as practical prediction systems.
Solar Irradiance Data
Time series change patterns of recorded solar irradiance
Extracted Rules
Prediction rules expressed in IF-THEN format
Developed Software Examples
We introduce software tools developed for time series data analysis.
Time series data Searcher
Software for visual and intuitive searching of time series data.
- CSV format time series data input support
- Intuitive operation with GUI
- Visual pattern search
- Pattern matching using Dynamic Time Warping
Query Dialog
Visually describe the pattern to search
Result Dialog
Display search results and CSV export
Time series data Analyzer
An integrated tool for searching and analyzing time series data.
- Flexible search with custom query language
- Machine learning with Weka engine
- Platform-independent Java implementation
- Automatic database creation functionality
Main Window
CSV file loading and database creation
Query Dialog
Specify combinations of multiple search patterns
Search Results Display
Visual confirmation of search results
Future Prediction Decision Tree
Future prediction results using machine learning