Application of improved clustering algorithm in precision poverty auditing - an example from region D

In order to accurately improve the audit of poor households in poverty alleviation projects, as well as the unreasonable parameter setting and low clustering accuracy of clustering analysis algorithm in its application. The study proposes a KLS-DBSCAN cluster analysis algorithm. The algorithm first uses kernel function estimation to determine a reasonable interval for the neighbourhood and the minimum number of nodes then uses the data local density characteristics to determine the number of clusters according to the parameter values within the reasonable interval, followed by the maximum contour coefficient to determine the optimal parameters. The optimal combination of hyperparameters for the KLS-DBSCAN cluster analysis algorithm is (0.25, 3), with 42 outlier points and nine clusters. Compared with the other three clustering analysis algorithms, the number of outliers in clusters is about 20. This research providing possibilities and technical support for the proper implementation of precision poverty alleviation audit work.