Performance Comparison of Neural Networks: A Case of Data Scientists' Job Change Prediction

In today's world, the era of big data, companies in every sector have to deal with huge amounts of data generated. Such data must be processed, analyzed, and interpreted to be used in making business decisions. Businesses employ data scientists for this purpose. These people have great costs to businesses. For this reason, it is a significant issue for businesses to predict the employee who intends to change jobs in people working as data scientists in enterprises. In this study; the job change thoughts of data scientists were predicted by artificial neural networks. Data cleaning, missing data completion with linear regression-based iterativelmputer method, data balancing with SMOTE (Synthetic Minority Oversampling Technique) algorithm, data normalization with standard scaler method were performed on the dataset used, respectively. The dataset was then trained with a multilayer perceptron algorithm and a deep neural network model. The trained models were tested and an accuracy of 84.2% was obtained with the multilayer perceptron algorithm and 87.5% with the deep neural network model. To compare the performance of artificial neural network models, analyses were performed with the frequently used Naive Bayes, Support Vector Machines, Decision Trees, Random Forests, Extra Trees, Gradient Boosting, and XGBoost algorithms. As a result of these tests, an accuracy of 91.1% was obtained with the XGBoost algorithm and performance metrics were presented.