A Heuristic Approach to Parameter Tuning in a SMOTE-Based Preprocessing Algorithm for Imbalanced Ordinal Classification

Abstract

Datasets are the backbone for data mining and knowledge engineering field to build an excellent classification model. However, the learning model usually follows a significantly biased distribution of classes. It is known as a class imbalance. The class imbalance problem exists in many real-time datasets. The main objective of this research is to fix an adaptive threshold based on class data and fit that data into models that can be understood and utilized by ordinal multiclass imbalanced scenario for improving the predictive accuracy. The methodology utilizes Laplace-Gauss based SMOTE method for synthesizing sophisticated objects of minority classes. Dynamic parameters are adapted for SMOTE algorithm by utilizing the underlying class information. On the whole, the dataset is divided into training and test data. Training dataset is updated with new synthetic patterns. The experimental analysis is performed on testing dataset to check the efficiency of the proposed methodology by comparing it with the existing methodology. The performance evaluation is conducted in terms of the measures called Mean Absolute Error (MAE), 154Maximum Mean Absolute Error (MMAE), Geometric Mean (GM), Kappa, and Average Accuracy. The Experimental results prove that the proposed methodology can produce authentic synthetic patterns than the existing method. The proposed method can synthesize the new effective patterns with the help of dynamic parameter setting. It upgrades the global precision and class-wise precision especially preserves rank order of the classes.