Performance in COVID-19 Mortality Prediction Using High-Dimensional Tabular Data

View the PDF file for the paper entitled Long Language Models for Classic Automated Language: Performance in Covid-19 death prediction

PDF view

a summary:This study compared the performance of automatic learning models (CMLS) and large LLMS models in the Covid-19 deaths using high-dimensional table data of 9,134 patients across four hospitals. Seven CML models, including XGBOOST and Random Forest (RF), were evaluated, along with eight LLMS, such as GPT-4 and Mistral-7B, which conducted a zero classification on the structured data that was converted into the text. In addition, Mistral-7B has been seized using QLora approach. XGBOOST and RF showed a superior performance between CMLS, where they achieved F1 from 0.87 and 0.83 for internal and external verification, respectively. GPT-4 LLM has led with a F1 score of 0.43, while the Mistral-7B formulation greatly improved its summons from 1 % to 79 %, which resulted in a stable F1 score of 0.74 while checking external health. Although LLMS showed a moderate performance in the classification of lead, the seizure dramatically enhances its effectiveness, which may block the gap with CML models. However, CMLS still outperforms LLMS in dealing with high -dimensional tabular data tasks. This study sheds light on the capabilities of both CMLS and LLMS seized in medical predictive modeling, emphasizing the current supremacy of CMLS to analyze organized data.