Track: Undergraduate Student Paper Competition

Abstract

Internal auditing is a fundamental process for quality assurance and proper compliance of different processes within a company. For this reason, it is also very important that companies can adequately plan its execution; but it often happens that companies do not have the necessary tools to make proper planning. One of the most important aspects to consider for effective planning is the calculation of the estimated time it takes to carry out the audits. This aspect is vital for proper time organization. If there is an ineffective or imprecise calculation, planning can be significantly affected and trigger different consequences for the company and its auditing processes. Recognizing the importance of this aspect, this project proposes the use of machine learning algorithms to predict the effort time (referred to as Actual Effort) that is required to carry out a certain number of audits in a period of two weeks (referred to as Sprint) in an IT services company. It is worth mentioning that in the early stages of the project it was intended to develop a model that predicts the Actual Effort for a single audit, but due to high volatility, inefficient results and at the request of the company, the approach of predicting the Actual Effort per Sprint was preferred. Different types of supervised models were implemented along the project such as Multiple Linear Regression (MLR), Random Forest Regressor (RFR), KNN Regressor, Support Vector Regression (SVR), Extra Trees Regressor (ETR), ANN Regressor, Gradient Boosting Regressor (GBR), among others. Various evaluation metrics were used to asses the models, such as coefficient of determination (R²), root mean square error (RMSE), mean square error (MSE), mean absolute percentage error (MAPE), symmetric mean absolute percentage error (SMAPE), and the mean arctangent percent error (MAAPE). Different tests were carried out with different versions of the database where the best results were obtained by an ETR model with an R² of 96% and a SMAPE of 8.47% were obtained.