Enhancing Prediction Accuracy through Normalization Techniques: Practical Insights from MLP Neural Network
Bashar Alhajahmad, Musa Ataş · 2023
Data normalization plays a pivotal role in enhancing the effectiveness of machine learning models across a myriad of applications. This article delves into the influence of various data normalization techniques on the performance of a Multilayer Perceptron (MLP) classifier, a prominent architecture in Artificial Neural Networks (ANN). The study explores the effects of three distinct normalization methods: Min-Max, Z-Score, and Decimal Scaling, supported by the Knime environment. The impact of these normalization techniques on model performance is investigated using a dataset sourced from social networking sites, containing user information relevant to a new SUV model's potential purchase. The research employs a visual representation of confusion matrices and ROC Curve charts to assess the outcomes of each normalization technique. Findings reveal that without normalization, the model's performance is suboptimal, with low accuracy and high error rates. In contrast, the application of normalization techniques significantly improves model performance. Notably, Z-Score normalization emerges as the most effective, achieving an impressive 92.5% accuracy with a reduced error rate. This study underscores the critical role of data normalization in machine learning and provides valuable insights into the choice of appropriate techniques for different datasets. Such insights can lead to more robust and reliable results, ultimately enhancing the quality and reliability of machine learning applications.