Optimization of Skewed Data Using Sampling-Based Preprocessing Approach.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• Data classification faces challenges due to the increasing volume and uneven distribution of data across classes.
• Traditional machine learning algorithms often neglect minority classes, leading to skewed results and reduced classification accuracy.
• The data-skewing problem is a critical issue requiring effective preprocessing strategies.

Purpose of the Study:

• To investigate the effectiveness of data preprocessing using sampling techniques to mitigate the data-skewing problem in classification tasks.
• To compare the performance of different sampling methods in addressing uneven data distribution.
• To evaluate the impact of these techniques on the efficiency of the K-nearest neighbor classification algorithm.

Main Methods:

• Implementation of three distinct sampling techniques: Resampling, SpreadSubSampling, and Synthetic Minority Over-sampling Technique (SMOTE).
• Application of these sampling methods to address data imbalance in classification datasets.
• Classification of preprocessed data using the K-nearest neighbor (KNN) algorithm.
• Performance evaluation using various classification metrics to assess efficiency.

Main Results:

• Sampling techniques significantly reduce data-skewing, improving the representation of minority classes.
• SMOTE and other methods demonstrate enhanced classification performance compared to baseline approaches.
• The K-nearest neighbor algorithm shows improved accuracy and efficiency when applied to data preprocessed with sampling techniques.

Conclusions:

• Data preprocessing with sampling techniques is crucial for overcoming data-skewing in machine learning classification.
• SMOTE and similar methods offer effective solutions for handling imbalanced datasets.
• These approaches enhance the reliability and accuracy of classification models, particularly in scenarios with uneven data distribution.