LVQ人工神经网络在伤寒、副伤寒发生强度判别与预测中的应用

Application of LVQ artificial neural network in discrimination and forecasting of occurrent intensity of typhoid and paratyphoid

目的: 探讨学习矢量量化(LVQ)人工神经网络在伤寒、副伤寒发生强度判别与预测中的应用。方法:以前一年的平均气压、平均气温、平均降水量和平均蒸发量4个气象指标的标准化后的变量及伤寒、副伤寒发病率平方根反正弦变换值为研究自变量,将1979-2000年辽宁省某市伤寒、副伤寒发病率按大小分为高、中、低3种情况进行判别与预测研究。利用软件MATLAB6. 5的人工神经网络工具箱分别进行LVQ人工神经网络的构建、训练与模拟,分别考察LVQ人工神经网络在模型拟合及前瞻性和回顾性预测方面的能力,并且与传统Bayes判别分析进行比较。结果: LVQ人工神经网络能够从另一个角度对数据进行分类判别与预测,利用1980-1995年数据拟合准确率为100%,预测1996-2000年发病强度准确度为3 /5;利用1982 -2000年数据拟合准确率为100%,预测1 9 8 0 -1 9 8 1年发病强度准确度为1 /2,均略高于传统Bayes判别分析。随机选择1 6年数据的拟合准确率为93. 8%,预测另外5年发病强度准确度为4 /5,与传统Bayes判别分析相当。结论: LVQ人工神经网络能够与传统Bayes判别分析相媲美,在发病率预测方面具有广阔应用前景。

Objective: To investigate the potential of learning vector quantization (LVQ )artificial neural network tools for discrimination and forecasting of occurrent intensity of typhoid and paratyphoid. Methods: The independent variables included standardized variables of average air pressure, average air temperature, average precipitation, amount of evaporation, and square root arcsine transformed value of incidence rate. The incidence rates of typhoid and paratyphoid were divided into high, moderate, and low. Software MATLAB 6.5 was used to structure, train, and simulate the LVQ artificial neural network. The power of prediction on fitting, prospection, retrospection was investigated separately. The results were compared with traditional Bayes discriminant analysis. Results: The prospective model yielded accuracy of 100% based on data from 1980 to 1995 and accuracy based on data from 1996 to 2000 reached 3/5.The retracing model yielded accuracy of 100% based on data from 1982 to 2000 and accuracy based on data from 1980 to 1981 reached 1/2. The random model yielded accuracy of 93.8% based on data of 16 years randomly and accuracy on data of the other 5 years reached 4/5. The LVQ artificial neural network yielded excellent discrimination and calibration which were better than or equal to that of Bayes discriminant analysis. Conclusion: The LVQ artificial neural network provides useful information in the discrimination and forecasting of incidence rate of typhoid and paratyphoid and has a good applied future.