基于粒子群-支持向量机的时间序列分类诊断模型

Time Series Classification Diagnosis Model based on Partical Swarm Optimization and Support Vector Machine

摘要:

构建一种基于粒子群算法支持向量机(PSOSVM)的磁共振功能成像(fMRI)时间序列分类诊断模型，通过针对脑区多维时间序列数据的深层次分析实现病症患者和健康者的准确判断与区分，为面向fMRI时间序列数据的病症诊断和预测提供有效科学依据.该方法在以下4个方面不同于其他已有相关研究工作：(1) 构建基于自回归模型的脑区多维时间序列数据特征表示；(2) 构建基于支持向量机模型的脑区多维时间序列数据分类机制；(3) 构建基于粒子群算法的分类学习参数寻优策略；(4) 建立融合上述特征表示、优化分类与参数优选模式的fMRI时间序列数据分类诊断模型.通过以精神抑郁症作为实证分析的具体案例，所提出分类诊断模型已取得良好实验效果，展示出其有效性与合理性.

Abstract:

This paper presents a fMRI timeseries classification diagnosis model based on particle swarm optimizationsupport vector machine (PSOSVM), which achieves more accurate judgments and distinctions between patients and healthy individuals by deeply analyzing multidimensional timeseries data of brain regions. This approach is significantly different from the other existing related research work in four aspects as follows: Constructing the feature representation for multidimensional timeseries data of brain regions based on the autoregressive (AR) model; Constructing the classification scheme for multidimensional timeseries of brain regions based on the support vector machine (SVM) model; Constructing the parameter optimization strategy for the classification learning based on the particle swarm optimization (PSO) algorithm; Constructing the classification diagnosis framework for fMRI timeseries data by integrating the above feature representation, optimized classification and parameter optimization patterns. With the mental depression disorder (MDD) as a specific case of empirical analysis, our classification diagnosis model has obtained very positive numerical computation results.