Breast cancer is one of the most frequent and having a high mortality rate among women. Early detection of this enhances survival rates from 56% to more than 86%. As a result, an accurate and dependable approach is required. Predicting cancer survival is becoming an increasingly difficult task in medicine. Researchers have widely employed machine learning methods to address this difficulty. The k-nearest neighbour (KNN) technique is the most widely utilized among the various machine learning algorithms. This article examines the performance of various KNN versions (Classic one, adaptive, locally adaptive, k-means clustering, fuzzy, mutual, ensemble, Hassanat, and generalised mean distance) in predicting breast cancer survival. This study carried out massive implementations and experiments using Haberman's Survival Data Set, which was collected from Kaggle, to analyze these variants. For comparison analysis, we took into account the accuracy, precision, and recall performance metrics. Based on performance metrics, this study determines that the Hassanat KNN version performed the best, followed by the ensemble approach KNN. Based on four performance criteria (Accuracy, F1 score, Precision, and Recall) for survival prediction, the presented research summarizes which KNN variation is the most promising candidate to pursue. The results of this study could be utilized by beneficiaries and healthcare researchers to choose the best KNN variation, for predicting breast cancer survival.