基于代表率与投票机制的分类方法监测片剂包衣终点的研究

利用近红外技术,建立了一种基于代表率和投票机制的分类方法监测片剂包衣终点。首先,采用系统聚类对各包衣时间点样本均值进行聚类,建立分类模型;然后,用本文提出的代表率对各时间点样本进行模型验证,通过代表率验证均值聚类模型,明确模型的有效性和可行性;最后,用训练模型预测各时间点样本类别,通过投票机制获得各时间点最终投票结果,弥补了系统聚类无模型预测的不足,也可避免个别样本导致的错误判断。经实验数据验证:代表率最低的是T100M,为73.33%;最高的是T130M,为100.00%;平均代表率为86.77%。投票结果得出第一批测试数据和第二批测试数据的包衣终点分别为1T130M、2T132M,另外投票机制还能监测其他时间点的包衣过程。

Study on monitoring the end points of tablet coating by classification method based on representative rate and voting mechanism

A classification method based on representative rate and voting mechanism was established to monitor the end point of tablet coating by using near-infrared(NIR)technology. Firstly, the sample mean of each coating time point was clustered by hierarchical clustering, and establishing the classification model. Then, the representative rate proposed in this paper was used to verify the model at each time point. The method verified the mean clustering model through the representative rate to confirm the availability and feasibility of the model. Finally, the training model was used to predict the sample categories at each time point, and the final voting results at each time point were obtained through voting mechanism, which made up for the lack of model prediction in systematic clustering and avoided the wrong judgment caused by individual samples. The experimental data verified that the lowest representative rate was 73.33% for T100M, and the highest was 100.00% for T130M, with an average representative rate of 86.77%, The voting results show that the finishing points of the first batch of test data and the second batch of test data are 1T130M and 2T132M respectively. The voting mechanism can also monitor the process of coating at other time points.