Intelligent under-sampled data classification algorithm based on machine learning and KNN

A data classification algorithm based on machine learning and k-nearest neighbor intelligent undersampling is proposed. The proposed algorithm targets the challenges posed by an excess of majority-class samples and the scarcity of minority-class instances in imbalanced classification tasks, which leads to severe information loss and insufficient classification accuracy in traditional undersampling methods. First, a k-nearest neighbor intelligent under-sampling method based on density awareness and boundary recognition is designed. By improving the k-nearest neighbor calculation model, redundant samples in the majority class are accurately screened and key boundary samples are retained to achieve data reconstruction. Secondly, by integrating the Bagging ensemble strategy with the support vector machine classifier, the parallel ensemble feature of Bagging is utilized to alleviate the information bias caused by under-sampling, and the high-dimensional classification advantage of the support vector machine is combined to improve the classification performance of imbalanced data. The findings demonstrate that each of the three undersampling methods ultimately retains 34 majority class samples to achieve class balance, and the retention rate of boundary samples in random undersampling is only 44.68%, with a large number of key boundary samples being mistakenly deleted. The proposed k-nearest neighbor intelligent undersampling algorithm has a boundary sample retention rate of up to 95.74%, a redundant sample elimination rate of up to 92.36%, and a data information retention rate of up to 0.94, proving that this method can accurately distinguish boundary samples from internal redundant samples and effectively solve the defects of information loss and boundary blurring in traditional undersampling. The proposed algorithm can effectively balance the classification accuracy of the majority and minority classes, reduce information loss, achieve intelligent screening of imbalanced sample data and precise retention of boundary samples, and solve the problem of insufficient generalization ability of a single classifier on imbalanced data.