Prediction of Postpartum Hemorrhage by Adaptive K-Nearest Neighbor Based on Influence Factors

For the binary classification problems that the available values for all feature attributesin the dataset are discrete conventional, K-Nearest Neighbor(KNN) considers that the distances between different feature attribute values are the same when calculating the distance between the predicted sample and the training sample, which is not consistent with the actual situation. In this paper, we proposed an algorithm named Adaptive K-Nearest Neighbor Based on Influence Factors (AKNN-IF) which solved the disadvantages of KNN algorithm in calculating the distance of discrete conventional feature attributes. Moreover, as the size of the dataset grows, the influence factors will reflect the degree of the influence more realistically of different feature attribute values on label attributes and it's more effective to use influence factors to calculate the distance of different feature attribute values. Meanwhile, the impact of imbalanced data on the classification effect is weakened. As the influence factor is obtained through self-learning of the data, no human intervention is required, so that the algorithm has a certain self-adaptability. We tested some data which contained 1829 samples and proved that the classification effect of the AKNNIF algorithm is superior to the KNN algorithm, SVM, and C4.5 for the binary classification problems in which all the feature attributes are discrete conventional. AKNN-IF can contribute to the prevention and control of postpartum hemorrhage.