基于随机森林与最大相关最小冗余的泥石流信号识别模型

Debris Flow Signal Identification Model Based on Random Forest and Maximum Relevance Minimum Redundancy

摘要: 本文基于最大相关最小冗余(mRMR)准则与随机森林算法，提出一种具有数学可解释性的泥石流信号识别模型。以瑞士Illgraben地区地震信号数据为基础，从时域、频域及时频域提取多维特征，并基于互信息理论构建mRMR特征筛选模型，通过优化对应目标函数遴选出5个最具判别力且冗余度最低的特征。利用该特征子集训练随机森林模型，依据基尼不纯度最小化原则生成决策树，并通过集成投票机制实现分类。结果表明，所建模型在准确率与AUC (Area Under the Curve)方面均优于传统机器学习方法，仅用5个特征即达到与全特征随机森林相当的识别性能，且在6场独立事件测试中全部正确分类。本研究不仅提供了有效的泥石流识别工具，也从特征约简与集成学习耦合的角度为地质灾害信号处理建立了数学模型支撑。

Abstract: This paper proposes a mathematically interpretable debris flow signal identification model based on the Maximum Relevance and Minimum Redundancy (mRMR) criterion and the Random Forest algorithm. Using seismic signal data from the Illgraben area in Switzerland, multi-dimensional features are extracted from the time, frequency, and time-frequency domains. A feature selection model based on mutual information theory is constructed using mRMR, and by optimizing the corresponding objective function, the five most discriminative and least redundant features are selected. This feature subset is used to train a Random Forest model, where decision trees are generated based on the principle of Gini impurity minimization, and classification is achieved through an ensemble voting mechanism. The results show that the proposed model outperforms traditional machine learning methods in both accuracy and AUC, achieving recognition performance comparable to that of a full-feature Random Forest model using only five features, and correctly classifying all 6 independent event tests. This study not only provides an effective tool for debris flow identification but also establishes a mathematical model support for geological hazard signal processing from the perspective of coupling feature reduction and ensemble learning.