Improving Machine Leaming Accuracy using Data Augmentation in Recruitment Recommendation Process

Socio-economic and technological developments have changed the way companies manage Human Resources. Artificial Intelligence applications, such as Machine Learning (ML), can be applied as a recommendation system or decision-making system used in the recruitment process. The problem is that the dataset available to create this employee classification model is not always large enough. Moreover, the development of an exceptional predictive model in ML requires a reasonable dataset size. In this study, we implemented Synthetic Minority Oversampling (SMOTE) as an oversampling technique while taking the basic concept as a data augmentation technique to increase the number of samples. We performed the data oversampling and augmentation using the k-NN-Linear Interpolation technique used in the SMOTE method. Besides being able to overcome the unbalance of datasets, this technique can also provide sufficient and representative data. The use of this augmentation technique can solve the problem of overfitting and result in high accuracy. According to the experiment, MultiLayer Perceptron (MLP) algorithm became the best performance model with a validation accuracy of 0.985. The best classification model was used to classify new employee candidates. The results obtained were recommendations in personality.