نوع المشرف

مشرف مشارك

تاريخ الاشراف على الرسالة من

29/04/2020

الى

22/06/2021

اسم الطالب

Dareen Ali Mohammad AbuRabi'e

ملخص الرسالة

Recruiting staff is one of the most difficult and important decisions to be made by the management. Hiring the wrong candidate would lead to losing valuable potential employees, while recruiting the wrong candidates may lead to wasting the organization resources, profit, and reputation. It may also expose the employer to troubles and may lead to legal procedures. In this work, we propose an intelligent approach for staff recruitment that employs machine learning, data mining, text mining and natural language processing (NLP) for performing smart staff recruitment. This work aims at enabling employers to utilize artificial intelligence techniques to perform unbiased, efficient, and smart automated recruitment of the best candidates which would help the organization to guarantee growth and prosperity. Our proposed approach involves employing data mining for finding the most important predictors of successful staff performance using the organization's historical data. A job specification is then automatically generated which includes recruitment criteria based on the identified predictors. Text mining and natural language processing are then applied to match the candidate’s CV to the job specification to screen and shortlist candidates. Our proposed system was applied to a dataset that was acquired from the Jordanian Department of Statistics (DOS) which consists of profiles of 529 employees that contain 18 features. The dataset was used for constructing 27 models that were generated in three experiments and using nine machine learning algorithms. The best performance was achieved using the K-Nearest Neighbours (KNN) which scored 91% classification accuracy, Random Forest with 89% classification accuracy, and Random Committee 86%. The results were excellent and were also better than most of the results that were reported ins similar studies. As for the results of CVs matching, the performance achieved was 80% using the random forests algorithm.