Comparative analysis of machine learning and survival analysis for breast cancer prediction
Madeline Emily, Felicia Meidioktaviana, Ghinaa Zain Nabiilah, Jurike V. Moniaga · Procedia Computer Science · 2024
Breast cancer cases are an alarming problem for several countries and are among other female diseases which are in the top list of population ill-health. To overcome this obstacle, it is crucial to develop reliable predictive models with high accuracy to maximize impact on early detection of breast cancer. Since we want to contrast two different models in studying patient death from breast cancer, our goal is to evaluate these models for their accuracy. The Cox model outperforms the Survival Random Forest in terms of survival probability, with a C-index of 0.757 and an AUC of 0.8561 compared to the Survival Random Forest's C-index and AUC of 0.5837. Furthermore, the Cox model's superiority is proved by its feature selection score of 46.72174, whereas the Survival Random Forest has no similar metrics. Additionally, Random Forest shows superior performance with higher precision in identifying survival probabilities with perfect specificity (100%), precision (100%), and overall accuracy (96.8%), versus 91.3%, 71.4%, and 83.9% for Naive Bayes. Random Forest is a more reliable solution for accurately anticipating both positive and negative cases due of its adaptability and zero false positive rate.