인공지능 기법을 활용한 공직기강 감찰활동 효과성 향상에 관한 연구

With the widespread digitalization, the availability and cases of RegTech are also increasing. Among them, cases proving that machine learning is highly likely to be applied to audit activities are drawing attention. In this study, we aim to improve the problems of post-detection-oriented public discipline inspection activities and utilize artificial intelligence technology to carry out preventive inspection activities on a regular basis. Specifically, the machine learning techniques were used to create a prediction model for PC ON overtime, and the possibility of using artificial intelligence techniques on a regular basis was studied for checking the absenteeism and tardiness of public office discipline. For this experiment, in-house PC ON time data of employees of a public institution were used from February 2020 to July 2020, and the entire data were separated into 75% of training data and 25% of test data, respectively, with using k-fold cross-validation method for improving prediction accuracy. For various predictive model generation experiments, five machine learning techniques were used: k-NN, decision tree, random forest, gradient boosting, and neural network. Optimization of each technique was performed to compare the accuracy of the optimization model. The results of the experiment confirmed that the ‘work type’, ‘employee number’, ‘department’, and ‘month’variables were selected as important characteristic elements when the PC ON overtime classification prediction model was created. The best model in the forecast was the optimized random forest model with 94.4% accuracy.