基于传播动态和集成学习的子种群重要性评估

Evaluating importance of subpopulations via transmission dynamics and ensemble learning

流行病的空间传播由复合种群网络中的重要子种群驱动,有效识别这些子种群对遏制疫情传播具有重要意义。为此,提出一种基于传播动态特性和集成学习的子种群重要性评估方法。对流行病在网络上的传播动态特性进行分析,在此基础上构建有效输出强度等能反映子种群传播能力的特征向量,使用集成学习算法XGBoost迭代训练评估子种群重要性的回归模型。所提方法在评估子种群重要性时,充分考虑网络结构和流行病增殖扩散的影响因素。实验结果表明,相比已有重要性评估方法,该方法能够更准确有效评估并区分子种群的重要性。

The large-scale spread of epidemics among different areas is often driven by a small number of influential subpopulations in metapopulation networks.Effective identification of these key subpopulations provides great help to contain the epidemic spreading.By analyzing the transmission dynamics of epidemics across the metapopulation network,the feature vectors such as effective output intensity reflecting the transmission ability of subpopulations were extracted.The subpopulation importance evaluation method was established using the ensemble model XGboost with the extracted features.In evaluating the importance of subpopulations,the topology of subpopulations was considered and the factors that influenced the spread of the epidemic were included.Experimental results show that the proposed method can identify and rank the importance of subpopulation more accurately and efficiently than other important evaluation methods.