基于带时间回溯的神经网络算法对新冠疫情数据的分析与预测

Analysis and Prediction on COVID-19 Data Based on the Neural NetworkAlgorithm with Time Backtracking

由于新冠肺炎疫情对全球造成了巨大的影响,因此有必要研究疫情的发展趋势。为此,笔者构建了带有时间回溯的神经网络模型,对以新冠肺炎累计病例为代表的时间序列数据进行分析和预测,首先通过K-means聚类方法对209个国家和地区进行分类,然后从不同类别中挑选代表国家或地区,应用改进的神经网络模型进行分析和预测,并和其他经典算法进行比较。实验结果证明,与其他经典的机器学习算法相比,经过改良的具有时间回溯的神经网络算法表现更好,预测准确度较高,能够有效预测新冠肺炎疫情的发展趋势及众多的时间序列数据。

As the COVID-19 epidemic has had a huge impact on the world,it is necessary to study the development trend of the epidemic.To explore the hidden trend of COVID-19,we build a neural network model with time backtracking to further analyze and predict the time series data(e.g.the cumulative cases of COVID-19).The K-means clustering method was firstly applied to classify 209 countries and regions into different categories,from which the representative countries or regions are selected.Then,the improved neural network model is proposed for analysis and prediction,and validated by the comparison with other classical algorithms.The 209 countries and territories were clustered into two categories,from which China and the United States were selected as representatives for analysis and prediction.According to the prediction results,our improved neural network algorithm with time backtracking performs better in both cases compared with other algorithms and has high prediction accuracy,which can effectively predict the hidden trend of COVID-19 and numerous time series data.