Transparency Prediction of Fraud Violations as an Anti-corruption Culture: Experiment of Decision Tree

Abstract

Several prominent reports have highlighted the unsatisfactory level of anti-corruption transparency for the private sector in Indonesia. Hence, the anti-corruption vision is still an aspect that deserves to be campaigned for to form an advanced and just civilization. This study aims to obtain a pattern of knowledge in predicting the level of transparency of disclosure of fraud violations based on a data mining approach. The classification function algorithm in this study is a decision tree which is then compared with other classification function algorithms, naive Bayes, and k-nn. The sample in this study is 141 companies combined in the construction, mining, and banking sectors, which are on the IDX for the 2019 period. As a result, the decision tree algorithm provides the second-best performance in predicting the level of corporate transparency, namely an accuracy of 70.92% and an AUC level of 0.740. Then in terms of different tests, the decision tree algorithm is in the same cluster as the algorithm with the best performance because the t-test results show no significant difference. Based on the pattern generated by the decision tree algorithm, the elements of opportunity, pressure, and arrogance are considered key factors in predicting the level of transparency of disclosure of fraud violations. One of them can be interpreted that a company that is supervised by a minimum of four independent commissioners means company tends to be predicted to be more daring in disclosing anti-corruption information in its annual report to the wider public. This study also recommends that every authorized institution in Indonesia can apply a data mining algorithm approach in utilizing the advantages of each agency's internal data volume to map anti-corruption cultural socialization strategies in private sector companies.