Comparison of Various Machine Learning Techniques Based on Variable Selection under Imbalanced Data

  Mo Ahsan Ahmad,  Umme Kulsum,  Faizan Ansari,  Mo Nafees,  Ravindranath Sawane

Classification in an imbalanced dataset is one of the challenges in statistical learning because many algorithms are designed to optimize overall accuracy without considering the relative class distribution. These algorithms are biased towards the majority class and tend to ignore the minority class, which is the class of interest for experimenters. This paper reviewed the data-level approach, algorithmic-level approach, and performance evaluation metrics for the classification of imbalanced data. Oversampling, undersampling, and hybrid sampling techniques are discussed along with Decision Trees Classifier, K-NN, Naive Bayes, Logistic Regression, Support Vector Machines, and Random Forest Classifier. The effectiveness of data-level approaches combined with algorithmic-level approaches to improve classification performance and compare the various machine learning classification algorithms of a “credit card fraud detection” dataset. Random Forest Classifier achieved the highest F-score and lowest log loss with and without resampling techniques.

Keywords- Imbalanced data, Variable Selection, Oversampling, Undersampling, Supervised Learning, Credit Card Fraud Detection

Unique Identification Number - IJEDR2204009
Page Number(s) - 60-70
Pubished in - Volume 10 | Issue 4 | December 2022
DOI (Digital Object Identifier) -    http://doi.one/10.1729/Journal.32282
Publisher - IJEDR (ISSN - 2321-9939)

  Mo Ahsan Ahmad,  Umme Kulsum,  Faizan Ansari,  Mo Nafees,  Ravindranath Sawane,   "Comparison of Various Machine Learning Techniques Based on Variable Selection under Imbalanced Data", International Journal of Engineering Development and Research (IJEDR), ISSN:2321-9939, Volume.10, Issue 4, pp.60-70, December 2022, Available at :http://www.ijedr.org/papers/IJEDR2204009.pdf