The application of deep learning algorithms for detecting anomalies in data is becoming increasingly prevalent, and accurate identification of these anomalies is of utmost importance. In this paper, the dimensionality of the data is augmented by employing a Gaussian kernel function during the data preprocessing stage, which allowed for the extraction of more potential features. To address the issue of poor prediction accuracy associated with random forest algorithms when faced with limited samples and features, adaptive algorithms based on random forest classifiers were utilized to propose a convergent anomaly detection model. Additionally, a comparative analysis of different combinations of water quality parameters is conducted to determine the optimal combination of indices.