An Enhanced Random Forest Algorithm Based on Spatial Partitioning and Weighted Voting Mechanism

Random Forest is an efficient machine learning classification algorithm known for its high accuracy and interpretability. However, the traditional voting mechanism in the standard Random Forest algorithm assigns equal voting power to all decision trees, regardless of their training accuracy, which can reduce overall model performance. To address this issue without overemphasizing individual trees through direct weighting, this paper proposes an enhanced Random Forest algorithm based on spatial partitioning and a weighted voting mechanism. The proposed method partitions the entire feature space, and within each subspace, a relative majority voting mechanism is applied. Additionally, a validation set is used to assign weights to each subspace, taking into account the differences between strong and weak classifiers. Experimental results on publicly available datasets demonstrate that the proposed algorithm outperforms the traditional Random Forest, SVM, and KNN classifiers in terms of classification accuracy and recall.