基于RFE引导PSO启发式特征选择的高维结直肠癌癌前病变分类

Classification of High Dimensional Colorectal Precancerous Lesion Based on RFE-Guided PSO Inspired Feature Selection

摘要: 针对结肠镜数据中存在的高维、小样本以及特征冗余严重等问题，构建一种RFE引导的PSO启发式特征选择与分类框架。首先，用F检验与相关性分析对原始特征进行初筛，以降低维度并减少冗余信息；引入RFE生成特征重要性评分引导搜索过程，并采用一种改进的PSO启发式优化策略实现对特征子集的全局搜索，结合局部搜索机制对候选解进一步细化，同时在适应度函数中引入特征规模约束与稳定性约束，增强搜索能力并提高最优解的鲁棒性。最后，采用RBF核的支持向量机作为分类器，在五折交叉验证下评估其分类性能。实验结果表明，其Recall、Accuracy、F1-score及G-mean等多种评价指标均优于对比方法。

Abstract: To address the challenges of high dimensionality, small sample sizes, and severe feature redundancy in colonoscopy data, we propose an RFE-guided PSO heuristic framework for feature selection and classification. Firstly, the raw features are pre-screened using F-tests and correlation analysis to reduce dimensionality and eliminate redundant information. RFE is introduced to generate feature importance scores that guide the search process, whilst an improved PSO heuristic optimisation strategy is employed to perform a global search for feature subsets. This is combined with a local search mechanism to further refine candidate solutions. Additionally, feature scale and stability constraints are incorporated into the fitness function to enhance search capabilities and improve the robustness of the optimal solution. Finally, a radial basis function (RBF) kernel support vector machine is employed as the classifier, and its classification performance is evaluated under five-fold cross-validation. Experimental results demonstrate that various evaluation metrics, including Recall, Accuracy, F1-score and G-mean, outperform those of the comparison methods.