一种基于KNN算法的手写数字分类器的设计与实现

Design and implementation of a Handwritten Digit Classifier Based on KNN Algorithm

手写体数字识别一直是机器学习分类领域研究的热点,文章设计了一种基于KNN算法手写数字分类器模型,使用主流的机器学习库scikit-learn进行开发,在预处理阶段,将数字集拆分为训练集和测试集,采用基于权重和不考虑权重的KNN算法进行模型训练和分类预测,利用网格搜索法根据分类预测准确率进行优化调参,最后在scikit learn库提供的UCI-ML手写体数字集进行测试,结果表明,文章设计的基于KNN算法的手写数字分类器模型,能够较好的完成UCIML手写数字测试集的分类工作,基于权重KNN分类模型的分类准确率为98.89%,基于非权重的KNN分类模型分类准确率为99.17%,另外本文也对手写数字体数据集归一化进行了讨论,结果显示基于权重的KNN分类模型和非权重KNN分类模型在数据集进行归一化操作后分类预测准确率并未有明显提升。

Handwritten numeral recognition has always been a hot topic in the field ofmachine learning and classification.In this paper,a handwritten numeral classifier model based on KNN algorithm is designed,which is developed by using mainstream machine learning library scikit learn.In the preprocessing stage,the number set is divided into training set and test set,andKNNalgorithmbased onweight and not consideringweight is used formodel training and classification predictionThe grid searchmethod is used to optimize the parameters according to the accuracy of classification prediction.Finally,the parameters are adjusted in scikit The uci-ml handwritten numeral set provided by the learn library is tested.The results show that the handwritten numeral classifier model based on KNN algorithm designed in this paper can better complete the classification work of uci-ml handwritten numeral test set.The classification accuracy of the weighted KNN classification model is 98.89%,and that of the non weightKNN classification model is 99.17%.In addition,the handwritten numeral is also tested in this paper The results show that the accuracy of KNN classification model based on weight and non weight KNN classification model has not improved significantly after normalization operation.