Comparison of deep-learning and conventional machine learning algorithms for salary prediction

Abstract

Salary is an integral part of contemporary life. With the large-scale use of machine learning, it has become possible to predict salaries with machine learning. Previous researchers have used random forest algorithms to solve this problem, however, there is a research gap in using a neural network to solve this problem. Therefore, the research topic of this paper is to use convolutional neural networks (CNN) and datasets on Kaggle to predict salary. The research methodology of this paper is as follows. First, the Kaggle dataset is divided into the train-dataset and the test-dataset. After preprocessing the data, two kinds of features are obtained. The features will be transformed into two two-dimensional matrices. Next, the matrices were used to train two CNNs separately. These two CNNs will be connected together to get the predicted salary by fully connected layers and Relu activation functions. After training the CNN, the study called the test dataset to verify the accuracy of the model. Similarly, the study used a random forest model for prediction. Finally, the comparison of the two results showed which algorithm was better. The study found that the error rate of the CNN was 0.0732 and its variance was 0.1899. The error rate of the random forest was 0.2437 and its variance was 0.8285. From the results, CNN is better than random forest in terms of accuracy and stability. Therefore, using CNN for salary prediction has a high probability of getting better results.