Penggunaan Algoritma Stacking Ensemble Learning dalam Memprediksi Pengguna Enroll pada Aplikasi Fintech

Abstract

One of the difficulties of companies engaged in the sale of products/services is targeting the right offers to potential users which will cause the funding used to attract users to not be carried out optimally. Predicting users who have the potential to enroll can be a reference for marketing implementation to be more optimal in future decision making, such as for example providing attractive promos to users who have the potential to not enroll to want to register and use the company's products/services. Making predictions involving machine learning has been done a lot, because machine learning can make decisions independently, this decision is made because machines can learn and recognize patterns from existing datasets. This technology requires algorithms in the learning process. Many studies have been carried out to improve machine learning performance by combining several algorithms (Ensemble Learning), random forest and adaboost are examples of combining several similar algorithms (homogeneous), but there is also a technique that combines several different algorithms (heterogeneous), namely stacking. This study predicts users who have the potential to enroll or register using stacking ensemble learning, a combination of naive bayes algorithm, random forest as a base learner and KNN as a meta learner, this study also applies feature selection using information gain, data transformation using z-score and produces accuracy 76%