Evaluation of applicant quality for a recruitment company using machine learning

It is important to the recruitment company that only quality candidates, and the candidates most likely to receive an offer of employment get short-listed to go through the carriers selection process. It is this second phase of the recruitment process, the short-listing of potential candidates, that will be the focus of this thesis. Data has being gathered by the recruitment company through an initial candidate screening process as well as responses from carriers regarding the selection outcome for each candidate. The topic of this thesis is to apply and evaluate machine learning algorithms to this data in order to predict which candidates are of the highest caliber and thereby most likely to receive an offer of employment from the carrier. The original screening data consisted of over 16 000 observations, However after removing observations with missing values and matching with the available response data the final data set consisted of only 1101 observations. One conjecture is that candidates with less experience or opportunities are more motivated to answer all screening questions thereby providing a more complete profile. Candidates with more experience, contacts and opportunities are less motivated to answer screening questions and thereby leave incomplete profiles, and consequently not making it into the final data set. The failure of the machine learning algorithms evaluated in this thesis to successfully classify candidates is not so much a failure of the algorithms to find a pattern in the data rather a reflection of the fact that there is a fundamental absence of a pattern in this particular data set. This does not mean that there are no patterns that could eventually be exploited to create accurate machine learning models, rather that improvements in the candidate screening and data gathering process are needed first.