The use of an adaptive distance measure for breast cancer treatments

摘要

Breast cancer is one of the leading causes of death among middle-aged and old women. Treatment decision-making may depend upon defined extent of disease, but it requires the knowledge of several other factors from patient and medical diagnosis. The measurement variability in some factors leads to the data with lots of noise. Most classification algorithms are very sensitive to noisy training data. The nearest-neighbor is a simple classification algorithm that is known to be very sensitive to the quality of the training data. In this paper, we use an adaptive distance measure for nearest-neighbor algorithm designed for noisy data to tackle the problem of classifying breast cancer treatments. This algorithm is based on assigning a weight to each training example. The weight assigned to a training example controls the influence of that example in classifying test patterns. The weights of training examples are assigned in such a way to minimize the leave-one-out classification error-rate on training data. To assess the performance of this method, we used clinical data about breast cancer treatments from 330 cases in an attempt to classify the treatment decisions. The results indicate that the proposed method can significantly outperform other methods proposed in the past for the task of classifying treatment decisions.