I have, since 1999, been working on scientific problems that can be broadly described as machine learning techniques for data analysis. In my thesis, I suggested two novel data mining algorithms based on Genetic Programming. I have contributed with theoretical results, methods, systems, algorithms and applications within the fields of data mining, soft computing and machine learning. Some key results include:

• An algorithm (G-REX) for rule extraction from opaque models
• Several algorithms for ensemble creation; e.g., GEMS
• An algorithm (Chipper) for rule learning
• Novel algorithms based on lazy learning (BuLL and G-kNN)
• Novel algorithms based on machine learning and soft computing techniques utilizing and extending the conformal prediction framework
• A novel methodology, named oracle coaching, used for building interpretable yet accurate models
• A novel methodology, based on a suggested concept named imaginary ensembles, used for selecting a specific classifier from a large pool
• Methods and theoretical findings regarding neural networks
• Methods and theoretical findings regarding ensembles
• Methods and theoretical findings regarding evolutionary computation
• Methods and theoretical findings regarding the concept description task
• Methods and theoretical findings regarding conformal prediction
• Theoretical findings and new criteria suggested for the rule extraction task
• Theoretical findings regarding the relationship between accuracy and diversity in ensembles
• Novel micro-techniques suggested within the fields of neural networks, genetic programming, ensembles, rule learning, lazy learning, feature selection and time series forecasting
• Applications within, for instance, drug discovery, health science, marketing, high-frequency trading, game AI, sales forecasting and gambling

Lyssna på Ulf Johansson i Retailpodden

Publications

My publications can be found at Google Scholar

Uppdrag

• Vetenskaplig ledare inom SIIR

Till forskarens publikationer i DiVA (Digitala Vetenskapliga Arkivet)

Avhandlingstitel

Obtaining Accurate and Comprehensible Data Mining Models : An Evolutionary Approach