Tuning Data Preprocessing Techniques in Enhancing the Accuracy of Self-Organizing MAP (SOM) for Prostate Genomic Cancer

This research investigates the impact of various preprocessing methods on the accuracy and effectiveness of Self-Organizing Map (SOM) clustering, specifically in the context of high-dimensional genomic data related to prostate cancer. The primary goal is to identify the most effective preprocessing technique to enhance the precision of SOM clustering. Key preprocessing methods examined include standard scaling, min-max scaling, log transformation, quantile transformation, and fuzzy triangular numbers. Clustering performance was evaluated using metrics such as quantization error, topographic error, Silhouette score, Davies-Bouldin index, unified distance matrix (U-matrix), learning rate, radius neighborhood analysis, and training time. The findings consistently demonstrate that the fuzzy preprocessing technique outperforms all other methods across these metrics, while raw (unprocessed) data performs the worst. The fuzzy method preserves key data characteristics, maintains neighborhood relationships, generates well-defined clusters, and accelerates SOM convergence. These results have practical implications in several domains. In the context of prostate cancer research, the use of fuzzy preprocessing can improve the accuracy of clustering algorithms, potentially leading to more precise identification of cancer subtypes or better-informed treatment decisions. Beyond genomics, this technique offers a robust solution for analyzing high-dimensional data in diverse fields such as bioinformatics, material science, and finance, where accurate clustering is essential. The findings also provide a practical guide for data scientists, offering insights into optimal preprocessing choices that can significantly improve machine learning model performance.