Drug Audit Based on Different Models and Different Feature Extraction Methods

Abstract

The audit data has the characteristics of large amount and complex form, and the current drug audit mostly rely on database methods, resulting in low audit efficiency. Although k-means and other clustering algorithms have been studied in the field of auditing, there are problems of poor clustering effect and low accuracy. To accelerate the intellectualization of drug audit in the era of big data, this paper proposes six experimental schemes based on k-means clustering model, Random Forest classification model, Multi-Layer Perceptron classification model and two feature extraction methods to conduct drug expiration audit on a hospital's drug delivery data in 2019. Then the most suitable model and feature extraction method for drug expiration audit are found by comparing relevant performance evaluation indexes. This paper first verifies the feasibility of experimental schemes by using the Titanic dataset, because it is also a binary-classification problem and contains both character and numeric features, which is very similar to the drug delivery data, which will enlighten the processing and feature extraction of drug delivery data. Finally, the effectiveness and feasibility of the models proposed in this paper are validated through specific experimental schemes.