Evidence-Based Process in Research Involving Machine Learning Algorithms

Abstract

The current state of research involving the application of machine learning (ML) algorithms on various topics that directly impact human beings does not sufficiently focus on identifying and filling in research gaps. Evidence is lacking that the maximum accuracy showcased in multiple research papers is the best accuracy possible. However, it is vital to fill in research gaps to be cost-effective and actionable in the big picture. This study proposes a guideline that can be useful for future work involving ML algorithms on high-risk topics to fill in research gaps as much as possible for a particular problem. In this study, 10 different models were conducted with 12.7 million different parameter combinations to create this guideline. The results from the experiments demonstrated experimentation with different algorithms and parameters is crucial in research to deduct which algorithm performs well. Exposure to accurate and inaccurate models can assist researchers and relevant professionals in highlighting evidence-based methods that might contribute to improved findings in various areas. It suggests what works and does not and which task is the most appropriate to fill research gaps. It is also important to consider the guideline produced in this study as inspiration. More research is necessary to improve the guideline.

Ähnliche Arbeiten