Balancing data with SMOTE variants using supervised machine learning algorithms to predict churn rate.

Abstract

[ES] Balancing problem New alternatives to obtain balanced data has been used in the recent times of machine learning applications. Oversampling and undersampling in data analysis are techniques used to adjust the class distribution of a source data. Presenting our data The dataset used in this project contains the churn rate of an e-commerce site, being this the of clients that would potentially cancel their membership. The situation is that a small amount of the samples contains the clients that are exited, thus, we are presented with an imbalanced dataset. SMOTE Stage In this project Multiple variants of SMOTE (Synthetic Minority Over-sampling Technique) are presented in a competition to find the model that find the best solution for this problem. We will select the algorithms for this task relying on the Smote Variants package that could be found in Python. Prediction Stage By means of a competition of widely used supervised algorithms we will obtain the optimal model to predict this churn rate.

[EN] Balancing problem New alternatives to obtain balanced data has been used in the recent times of machine learning applications. Oversampling and undersampling in data analysis are techniques used to adjust the class distribution of a source data. Presenting our data The dataset used in this project contains the churn rate of an e-commerce site, being this the of clients that would potentially cancel their membership. The situation is that a small amount of the samples contains the clients that are exited, thus, we are presented with an imbalanced dataset. SMOTE Stage In this project Multiple variants of SMOTE (Synthetic Minority Over-sampling Technique) are presented in a competition to find the model that find the best solution for this problem. We will select the algorithms for this task relying on the Smote Variants package that could be found in Python. Prediction Stage By means of a competition of widely used supervised algorithms we will obtain the optimal model to predict this churn rate.