Hybrid and conventional machine learning in comparison
19.04.2022 Faced with a shortage of data, the convential machine learning approach reaches its limits. In such cases, a hybrid approach may yield better results, according to a new study by the Institute for Intelligent Industrial Systems I3S of Bern University of Applied Sciences BFH.
In order to work, machine learning needs large amounts of data. However, if only small amounts of data are available due to limited resources such as sensors, time and budget, this can cause the traditional machine learning approach to perform poorly or even fail. In such cases, a hybrid approach combining various methods may yield better results, according to a recently published study by the Institute for Intelligent Industrial Systems I3S of Bern University of Applied Sciences BFH. In the study, additional physical knowledge in the form of formulas was applied to the dataset, which was shown to improve the performance of the algorithm. The study’s findings are particularly interesting in regard to digitalisation in industry, as there is often little data available in this context. The proposal for a larger-scale research project to further investigate the hybrid machine learning approach has already been submitted.
The article was published in the journal Applied Sciences of the Multidisciplinary Digital Publishing Institute (MDPI), a publisher of scientific open access journals.