A New Approach to Computationally-Successful Linear and Polynomial Regression Analytics of Large Data in Medicine

In the realm of healthcare, predictive modeling stands as a pivotal tool for deciphering patient outcomes and refining medical decision-making processes. However, the accuracy of machine learning algorithms, which underpin these predictive models, often falls short, leading to erroneous predictions. This study offers a new approach to optimize linear and polynomial regression models for healthcare analytics, which aims to tackle this challenge. In contrast to earlier efforts, this method focuses on using a scaled-down data transformation to improve linear regression model performance. The main goal of this study is to reduce the sum of squared errors (SSE) and improve the predictive power of linear regression models by using a data transformation function to reduce the size of all variables. In a series of experiments, we used non-Bayesian statistics in SPSS and Matlab to generate 40 trials of linear regression models, with 1,000 observations in each trial. In addition, we used SPSS for regression analysis, Excel for data manipulation, Wilcoxon signed-rank tests, and Cronbach’s alpha statistics for optimization model performance evaluation.Our findings show that the suggested scale-down transformation method is effective, since the sum of squared errors is significantly reduced (absolute Z-score=5.511, effect size=0.779, p-value<0.001, Wilcoxon signed-rank test). Furthermore, the optimized model's robust internal consistency was confirmed by inter-item reliability testing (Cronbach's alpha=0.993)