Internal fraud is a big problem for companies since it causes significant monetary losses. Several research studies have proposed to improve the personnel selection process using data mining. The present work suggests to use applicants’ historical information in order to predict if they will commit fraud during their working period in a company. There are models with high precision level but with a higher error rate to find fraud. After several ex­perimentations, around seven variables which contribute more to the model were found. Some of these variables match those mentioned in studies about antisocial personality disorder. The algorithm with best results was a convolutional neural network with 80% accuracy rate. It is concluded that applicants’ information is important to establish if they will commit internal fraud during their working period in a company.

Internal fraud is a big problem for companies since it causes significant monetary losses. Several research studies have proposed to improve the personnel selection process using data mining. The present work suggests to use applicants’ historical information in order to predict if they will commit fraud during their working period in a company. There are models with high precision level but with a higher error rate to find fraud. After several ex­perimentations, around seven variables which contribute more to the model were found. Some of these variables match those mentioned in studies about antisocial personality disorder. The algorithm with best results was a convolutional neural network with 80% accuracy rate. It is concluded that applicants’ information is important to establish if they will commit internal fraud during their working period in a company.

Internal fraud is a big issue for companies, resulting in relevant monetary losses. Several investigations have proposed improvements to the personnel selection process making using of Data Mining. The present work proposes to use past information of applicants to a company to predict if they will commit fraud during their stay. We find models with high precision, but that have a bigger classification error to find the fraud cases. After several experiments, we find around 13 features of this universe that are most relevant to the model. Some of these features match with features mentioned in literature about antisocial disorders. We conclude that there is value in applicant information to predict if they will commit internal fraud during their stay in the company.

Indexado en Latindex (Catálogo 2.0)