Optimizing the early glaucoma detection from visual fields by combining preprocessing techniques and ensemble classifier with selection strategies

Towards Explainable Machine Learning for Bank Churn Prediction Using Data Balancing and Ensemble-Based Methods

“…In their work, the SMOTE algorithm proved to be the most relevant for balancing the data. Concerning random under-sampling of the majority class, previous authors have shown that this is not desirable, especially in combination with the ensemble method, as it can lead to data information loss [21]. SMOTE's strategy is to create an artificial instance of a minority class through the following operating process: Considering an instance x i of the minority class, the algorithm starts by creating a new artificial instance from x i by first separating the k nearest neighbors to x i , from the minority class.…”

“…SMOTE's strategy is to create an artificial instance of a minority class through the following operating process: Considering an instance x i of the minority class, the algorithm starts by creating a new artificial instance from x i by first separating the k nearest neighbors to x i , from the minority class. Then, randomly choose a neighbor and finally generate a synthetic example on the fictive line joining x i and the selected neighbor [21,40,44,47]. This process is clearly described by Algorithm 1.…”

“…Ensemble-based methods consist of a combination of several independent basic classifiers that are in most cases decision trees (DT) but can also be artificial neural networks (ANN) or support vector machine (SVM), k-nearest neighbours (k-NN) or naive Bayes (NB) [21]. Each of these independent weak learners provides an alternative prediction of the whole problem and the final prediction results in a combination (usually by weighted or unweighted vote) of these alternative predictions [50].…”

“…A concrete example remains its best performance at the Netflix challenge [39] which has made ensemble methods very famous and highly recommended for the scientific community. Several research contributions have been done to improve the ensemble methods which currently remain a major challenger of deep learning which is most used [21]. Recently, many studies [40][41][42][43] have shown that the combination of ensemble models with preprocessing techniques improves performance in modeling of the unbalanced classification problem.…”

“…Standardization is an additional step in dealing with this problem but it has an adverse or favorable effect on the performance of certain algorithms that are said to be unstable to variable scales; knowing that a very large scale difference is more favourable for overfitting. Choosing ensemble methods overcomes this problem without the need to normalize the data [21]. The second contribution of this paper is to use the synthetic minority oversampling technique (SMOTE) method to balance the data while the third one is the building of stable and better performing models based on an ensemble approach.…”