Automated CV screening

• Main Author: Hauptfleisch, Mário Rivotti
• Publication Date: 2018
• Format: Master thesis
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: http://hdl.handle.net/10071/32236
• topic: Machine learning; Curriculum vitae

In this dissertation was predicted the outcome of job applications using machine learning. The focus of the outcome is on the first stage of the recruitment process, where there is a very large number of applications. The data used to train classifiers and predict targets was provided by a company’s human resources tool, where they receive structured online applications with their corresponding CVs. The applications are for different job positions in the same company. The first part the experiments is the preprocessing of the data. Here the reader can find a description of the variables and how they are processed with preprocessing techniques, to be used by the machine learning algorithms. In the second part you will find the results for the predictions made by the different algorithms. The ones that were used in this thesis are: Decision Tree, Random Forest, Gradient Boosted Tree, SVM and Artificial Neural Networks, all form the python’s sklearn package. The outcomes that are predicted in this work are if the candidate passed the first stage of the screening process, if the candidate failed the overall process, if the candidate was hired, and the grades given to the applications labeled by the hiring managers. Results showed that by using the variable "Job ID", that describes the job that each candidate is applying, improved the predictions significantly. Without using the ’Job ID’ the best accuracies achieved were around the 75%. Using the "Job ID", the best accuracies were around 90%. Overall the Random Forest and Gradient Boosted Tree had the best results. The attributes that contributed the most to predict the different targets were the specification and area of study, the highest education achieved, the number of languages spoken, and the distance from home to work.