Skip to Content


Science photo

Mainly my scientific activities are machine learning, artificial intelligence and systems theory, all of them deployed to various real life applications (from bankruptcy prediction models to emission reduction decision support software). The motivation for doing so is to create efficient computer tools supporting humans.

The theme of my PhD dissertation was "Application of Discrete Predicting Structures in an Early Warning Expert System for Financial Distress". Its main idea was to use Kalman filters to build a dynamical classifier and test its efficiency in financial distress forecasting. Previous researches in this field were limited to methods assuming static behaviour of crisis processes. My idea was to apply technical methods (UKF - a non-linear Kalman filter) to solve financial problem via creation of dynamic classifier (dynamic in terms of systems theory - processes with 'memory' or 'inertion').

The results of this new attempt were compared with result achieved using well known methods in this domain - Neural Networks, Discriminant Analysis, Nearest Neighbour, Nearest Mean. They proved that taking dynamic into account one can build more efficient classifier.

I have bankruptcy dataset available at this website (check 'usage' section). If you are working in field of data mining, AI, computer systems in management or financial analysis and have any cooperation proposals - simply mail me.

Important question often rises when science is involved - "what could we use it for?". I think that it ought to be a guideline for all researchers. No science only for pure science. Even if our researches are sometimes quite small or unfunded, still they can bring new quality to life. That's what I like most in science - a fun of creation.

One of the biggest problems in science if amount of data to process. Huge number of processed variables can cause efficiecy decrease instead and increase of complication ("curse of dimensionality"). If you are looking for feature selection algorithm I developed CWFS algorithm.