Self-organizing, Parallel, High-Dimensional Modeling and Improved Model Evaluation!

Features of (yX) for Excel version 2.2

• [new] Introduction of a targeted modeling purpose as part of the setup process: Prediction in time, static modeling, binary classification;
• [new] For static and binary classification models three different Active Neuron types are available now: Polynomial, Radial Basis Function, and Sigmoidal function. Full support for all three types including exporting models to Excel;

Features of (yX) for Excel version 2.1

• [new] Improved prediction accuracy when using Similar Patterns modeling technology;
• [new] For regression models, model export to Excel improved;
• [new] Prediction project added which models and predicts global land air and sea surface temperatures for 9 global areas 36 months ahead. This is an open project which will be updated regularly!

Features of (yX) for Excel version 2.0

• [new] High-dimensional modeling which scales to data sets of very small (< 20) to large number of samples (<= 1M) and from small to large number of potential inputs (<= 50,000). Not any prior variables selection is necessary!
• [new] Original tools to noise immunity and evaluation of models derived from data to improve reliability of models and to minimize the risk of getting invalid or chance models from badly sized data sets. More ...
• [new] 64-bit parallel software. Ultra-fast processing of very computation intensive algorithms by 64-bit, multi-core/multi-processor, and vector processing support. The software automatically scales to the number of CPU-cores found at runtime to take full advantage of current and future multi-core hardware. The basic rule: Modeling speed grows with every core and/or processor available at runtime. More ...
• [new] Climate change modeling and prediction project added. This project is possible now since it requires building reliable models from data sets which are high-dimensional in input variables space;
• [updated] Tutorial and documentation;

Features of recent (yX) for Excel versions

• [new] Improved model selection procedure during model self-organization which may lead to more accurate models;
• [new] Automatic generation of per-sample model and prediction uncertainty now also for self-organized regression models;
• [new] Automatic generation of graphs for models in Excel including a prediction uncertainty plot;
• [new] Significant performance improvement for building regression models on data sets with large number of samples;
• [new] Essential memory requirement reduction for self-organizing regression modeling;

Features of version X 6.0 (Mac OS X only)

• [new] First version which is bundeled with KnowledgeMiner (yX) for Excel parallel software, version 1.0;
• [new] Cost curves added for displaying optimized cost-sensitive classification models;
• [updated] Business Intelligence Case Study app version 1.0.8;
• [fixed] Class value assignment for classification models in model report;

Features of version X 5.4 (Mac OS X only)

• [new] Displays uncertainty interval in regression plots for cost-sensitive models; This makes interpretation of results easier.
• [new] Displays uncertainty interval in regression plots for cost-sensitive models; This makes interpretation of results easier.
• [new] Platinum edition now supports table dimensions of up to 30,000 rows and 30,000 columns;
• [new] Built-in cost-sensitive modeling for both classification and regression models according to a given cost-benefit matrix and an accepted target uncertainty; this allows self-organization of predictive, cost-optimized models;
• [new] Enhanced and extended AppleScript support for building models and their workflow integration;
• [updated] Tutorial and documentation;
• [updated] MS Excel model export v.1.1.2;
• [updated] TransformModel version 1.0.10;

Selected features of previous versions (Mac OS X only)

• [new] improved significance testing during model self-organization to avoid using non-relevant inputs in a model;
• [new] special time limited educational price for the Platinum version. Discover knowledge in your data and publish on our site;
• [new] automated fuzzy model based defuzzification implemented, which simplifies use of fuzzy models;
• [new] defuzzification model self-detects if related fuzzy models are available and takes their outputs for back-transforming results into the initial data space;
• [new] works on both static and dynamic fuzzy models;
• [new] article on the problem of modeling ecotoxicity added;
• [new] new article on commodity price prediction added to the publications area;
• [new] extended model evaluation features for Platinum users (works also in demo);
• [new] decomposition of the overall model error into bias, regression slope, and random disturbances contributions;
• [new] detection of worse described output values;
• [new] detection of high leverage input vectors in both modeling and prediction phases;
• [new] marks high leverage input vectors in line and scatter plots;
• [updated] improved Analog Complexing pattern recognition algorithm; detection and handling of constant target patterns;

General features of version 5.0 (all supported platforms)

• for the first time, integrated noise filtering characteristics for a second level, on-the-fly model validation; supports evaluation if a model reflects a causal relationship or if it just models noise;
• explicit definition of exogenous variables for creation of systems of equations;
• two new data mining algorithms: Analog Complexing based clustering and classification (n classes);
• extended table dimension for up to 30,000 rows;
• program-to-program communication via AppleScript (Mac only):
  • creating and predicting GMDH, Fuzzy, and AC models from within a script;
  • allows for knowledge discovery workflow processing;
  • executable sample script for demand prediction included: importing data from databases (FileMaker or AppleWorks) or spreadsheets (MS Excel), data preprocessing, data mining using different technologies, prediction of new data, combining of predictions, and returning a final prediction (high, mean, low) to the initial data source;
• ROC analysis to measure classification power of a model;
• 5 hours data mining service included;
• a new Platinum Edition introduced;

(read more...) (870k - PDF)

Included datasets and examples range from prediction of global temperature, stock market trends, medical diagnosis, failure of materials (like the Challenger Space Shuttle O-Ring), recognition of handwritten digits, Wine Recognition, national economy, to party affiliation in the US congress. The many examples included with KM show its power to work on human issues.

Download and try our free trial version

Also, several papers added to the documentation

• "Prediction of Wastewater Pre-precipitation Variables using Self-Organizing Networks"
  Conference paper Limassol, Cyprus, 2005
• "A GMDH Approach to Modelling Gibbsite Solubility in Bayer Process Liquors"
  Journal paper 2004
• "Does my model really reflect a causal relationship?"
  About noise filtering power of GMDH, do not believe in a model's closeness-of-fit, only, and introduction of a new model quality measure: Descriptive Power, 2002