Extreme learning machine prediction under high class imbalance in bioinformatics

DI PERSIA, L. E.; STEGMAYER, G.; RODRIGUEZ, TADEO; MILONE, D.H.

Cordoba

Jornada; Simposio latinoamericano de investigación de operaciones e inteligencia artificial (SLIOIA); 2017

SADIO-CLEI

Class imbalance in machine learning is when thereare significantly fewer training instances of one classin comparison to another one. In bioinformatics, thereis such a problem in the computational predictionof novel microRNA (miRNAs) within a full genome.The well-known precursors miRNA (pre-miRNA) areusually only a few in comparison to the hundreds ofthousands of potential candidates, which makes thistask a high class imbalance classification problem. Itis well-known that high class imbalance usually affectsany classical supervised machine learning classifier.Thus the imbalance must be explicitly considered.Extreme Learning Machine (ELM) is a supervisedartificial neural network model that has gained interestin the last years because of its high learning rateand performance. In this work, we propose a novelapproach to overcome the high class imbalance inpre-miRNAs prediction data in which ELMs are usedfor predicting good candidates to pre-miRNA, withoutneeding balanced data sets. Real datasets were usedfor validation of the proposal with several class imbalancelevels. The results obtained showed the superiorityof the ELM approach against very recent state-ofthe-artmethods in the same experimental conditions.