Enhancing diversity and robustness of clustering ensemble via reliability weighted measure

To solve the problem of hidden pattern recognition and high dimensional perception of geospatial sensor data, machine learning can build a model of the unknown relationship between associated data and characteristic variables. In recent years, clustering ensemble (CE) is a hot research area in machine learning, which has attracted much attention by combining multiple base clustering results into an ideal clustering solution and improving the robustness of clustering performance. To better solve the intelligent prediction of landslides in geological disasters, a novel CE method based on mountain monitoring text data is designed and implemented to analyze and predict landslide disasters. First, the effective information is extracted by preprocessing the original data. Then, the reliability weighting and weighted measure formulas were designed by combining the concepts of cosine similarity, information entropy and reliability. Accordingly, a novel clustering ensemble algorithm based on reliability weighted measure (RWMCE) was proposed to mine and analyze the effective information implicit in mountain monitoring text data. Furthermore, the method can reduce the attention to the poor clusters and improve the overall prediction performance by weighting from the cluster level and measuring the reliability of weighting. To test the diversity and robustness of RWMCE in the landslide monitoring data set, ten groups of schemes based on K-means, K-medoids, Fuzzy C-Means, and Gaussian Mixture Model were set up for performance evaluation. Experiments on nine publicly available datasets proves that RWMCE has better prediction performance for landslide geological disasters.