Advancements in Cancer Survival Prediction: A Systematic Review of Classical and Modern Approaches

The unpredictable nature of cancer, along with the absence of noticeable symptoms, makes it essential to accurately predict patient survival in order to enhance treatment results. Conventional approaches often struggle with the complexity of cancer. As digitization continues to grow, advanced machine learning and deep learning models are increasingly used to improve survival predictions. This paper aims to identify the survival analysis models applied in cancer prediction, highlight recent advancements, and suggest directions for future research. A literature search was conducted using three databases: ScienceDirect, IEEE Xplore, and PubMed. Boolean search strategies were used to locate relevant studies published in the last 15 years. The PRISMA guidelines were followed to review and select articles based on predefined inclusion criteria. This review critically examines 51 articles, focusing on the transition from traditional statistical methods to more advanced machine learning techniques. The findings show a growing trend towards using clinical data, even when the data sets are limited, and an increasing interest in hybrid and deep learning models for survival prediction. While traditional machine learning methods still hold a majority, the potential of deep learning and integrated techniques is gaining wider recognition. The findings emphasize the need for improved machine learning approaches to achieve more accurate survival predictions and encourage further research into deep learning models. It offers valuable insights for researchers at all levels, providing an overview of current methods and potential areas for future exploration in cancer survival analysis.