Accuracy prediction of identification in remote customer acquisition in banking with machine learning

Abstract

In banking, thanks to remote identity detection, the customer representative and the real person will not need to be physically in the same environment, and new customers will be reached quickly and effectively. In this study, a data set consisting of randomly generated contact information for remote identity detection methods that can be used in customer identity verification is trained using Natural Language Processing Techniques and an estimate is made as to whether the person is real or not. One of the methods used in this study is "Word Embedding". Word Embedding is a method for closely representing words with similar meanings. The generated data set is modeled with Word2Vec, a word vector algorithm. The clustering of the word vectors obtained by Word2Vec techniques, in terms of their formal properties as well as the semantic relations of the words they belong to, has been examined. Two different Word2Vec methods such as CBoW and Skip-Gram were used to create the model. According to the results of the application, a success rate of 89% was achieved in the estimation of the correct data.