Optimization of call center performance by call prediction using various machine learning methods

In recent years, when a problem is formulated as a prediction problem, literature offers various methodologies of ML and non-ML solutions. Thus prediction problems are highly studied, and especially in service systems comparative studies are suggested to find optimal solutions. In this study, the aim is to tidy up the existing recent literature about prediction problems in various fields and to utilise this literature to come up with a three option comparative study. A systematic literature review process has been followed, with 997 articles included into the process combined from queries from ScienceDirect, Scopus and Web of Science. Through elimination, 24 of them are fully analyzed. The results indicate that formulating the prediction as a classifier problem, which enables usage of XGBoost, Bayesian Classifier, regression etc. provides need for labeled data, and provides computational ease, whereas it can be less accurate, compared to deep learning solutions. Deep learning solutions are however "blackbox" as in in any field, literature cannot fully identify each iteration that has been recorded. Finally, literature suggests combining into hybrid solutions, as in using non-ML and ML solutions together, for example heuristic rulesets and deep learning to obtain a better fit.