Analyze and predict time series data with the ChaosKit library

ChaosKit is supplied as a web service, .Net or Java class libraries for storing, analyzing and predicting time series data using techniques from Chaos Theory. It contains a temporal database to store time-tagged samples of a single variable and can perform various well known analyses for chaotic behavior. It can calculate optimal embedding dimensions and separations, and perform short term iterated predictions on the data using a memory based modeling algorithm.

 Applications are:

• Analysis and prediction of complex systems with feedback
• Analysis of research data
• Quantitative analysis of financial time series data for short term trading of futures, stocks, bonds and currencies.

• Online evaluation environment
• Support for financial time series - data and predictions can be quantized into standard trading units and trading times/days can be specified.
• Calculates various Chaos Theory measures: Hurst Exponent, Lyapunov Exponent and Fractal Dimension.
• Uses Takens' embedding method to create a model of the attractor of the time series. 
• Determines optimal embedding parameters using the techniques of False Nearest Neighbors and Mutual Information.
• Memory-based prediction predicts from nearest historic neighboring points in the embedding space.
• Predictions can be single or multiple iterated values at user defined time intervals.
• An optional "forgetting" parameter ignores data older than a user configurable time.
• The class library includes support for general purpose KD Trees of arbitrary objects.