Comparison of Machine and Deep Learning Algorithms for Customer Transaction Prediction

There are enormous implementations of data analysis in business, and one of the various tasks in this area is customer transaction prediction. There are various algorithms for dealing with this task. That is why selecting an appropriate one from them is a difficult task. However, there is not enough research to give recommendations for algorithm selection. So, the research goal is to solve this problem for the company <<Santander>> by revealing the most effective practices for their further implementation into companies’ business decision-making process. For this purpose, Logistic Regression, KNN, Random Forest, Neural Networks, and their math parts were analyzed. Then they were implemented for the task solving with different optimizing approaches, hyperparameters, data preprocessing technics, and metrics. As the outcome of the experiments, the characteristics of the algorithms were defined. As the result of the research, the methodologies for customer transaction prediction were described, analyzed, and compared by their simplicity, interpretability, execution time, and accuracy. Then this comparison helped to prepare the final recommendations about model selection for this task to be implicated by analytics in their work to make it more responsible for business requirements.