Predict Employees’ Computer Access Needs in Company

When an employee is hired by a company, given his/her job role, model can be built to predict an employee’s access needs. This auto-access model tries to minimize the human labor and time for granting or revoking employee access. The input to our algorithm is information about the employee’s role at the time of approval. We then use Logistic Regression, Naive Bayes and Support Vector Machine (SVM) models to output a predicted ACTION for a specific resource. While, due to the unique characters of our data set, with only 6% samples labelled as 0, simply applying models like Logistic Regression, Naive Bayes and SVM can not get a satisfactory prediction on this unbalanced dataset. Alternatively, we use the One-hot Encoder to preprocess the data before applying different models and achieve a much higher accuracy (around 87%) in labelling computer access needs. Keywords—Unbalanced Prediction, Logistic Regression, SVM, Naive Bayes, One-hot Encoder, Cross Validation, Greedy Feature Selection.