Cancer: Investigating the impact of the implementation platform on machine learning models

In the context of global cancer prevalence and the imperative need to improve diagnostic efficiency, scientists have turned to machine learning (ML) techniques to expedite diagnosis processes. Although previous research has shown promising results in developing predictive models for faster cancer diagnosis, discrepancies in outcomes have emerged, even when employing the same dataset. This study addresses a critical question: does the choice of development platform for ML models impact their performance in cancer diagnosis? Utilizing the publicly available Wisconsin Diagnostic Breast Cancer (WDBC) dataset from the University of California, Irvine (UCI) to train four ML algorithms on two distinct platforms: Python SciKit-Learn and Knime Analytics. The algorithms’ performance was rigorously assessed and compared, with both platforms operating under their default configurations. The findings of this study underscore an impact of platform selection on ML model performance, emphasizing the need for thoughtful consideration when choosing a platform for predictive models’ development. Such a decision bears significant implications for model efficacy and, ultimately, patient outcomes in the healthcare industry. The source code (Python and Knime) and data for this study are made fully available through a public GitHub repository.