Author: https://github.com/halepino
A Power BI dashboard was created to examine relationships between employment variables and attrition (coded as yes 1/no 0). Recursive feature elimination (RFE) was conducted accross a range of N variables until an optimal amount was identified based on logistic regression scores on the test sample of employee data. Three models were then trained and compared for performance including a decision tree classifier, a gradient boosting classifier, and a mulit-layer preceptron (MLP) neural network. Following evaluation using classification reports, accuracy, and confusion matrices, the MLP model was found to have the best performance at 92% accuracy on test data.
Code Files(ipynb)
Power BI File (pbix)
White Paper (pdf)
Visual/Audio Presentation (pptx)
The data for this project was sourced from kaggle.com, here. The dataset included 1676 rows of employee's data with 34 columns. A complete list of variables is shown below.