Stable Distribution Naive Bayes Achieves Higher Accuracy Than Traditional Naive Bayes Classification

This study evaluates the performance of the Stable Distribution Naive Bayes Classifier on the well-known IRIS dataset, comparing it against the traditional Naive Bayes Classifier. The Stable Distribution Classifier, well-suited for data with heavy tails and skewness, consistently achieves superior accuracy, especially when handling outliers and nonstandard samples. This study conducted 18 feature combinations of Iris Versicolor and Iris Virginica across varying parameter configurations (𝛼, 𝛼), demonstrating the stable model’s robustness under constrained sample sizes. A significant technical contribution involves integrating R’s specialized stable package into Python, enabling the direct application of professional fitting and PDF functions for precise analysis. Representative results from key feature combinations further illustrate its practical advantages. Additionally, five additional datasets—Wine, Social Network Ads, Diabetes, Electrical Grid Stability Simulated, and Vehicle Silhouettes—further demonstrate the Stable Distribution Classifier’s broad applicability across diverse domains. This research further confirms that the Stable Distribution Naive Bayes Classifier is a robust and accessible alternative, offering enhanced predictive performance over models traditionally based on Gaussian distribution assumptions.