Neutrosophic Ensemble Classification for Uncertainty-Aware Bearing Fault Detection: Evidence from Laboratory and Variable-Speed Industrial Benchmarks

A neutrosophic ensemble of Random Forest, XGBoost and Logistic Regression reached 100.00 percent accuracy on three of four held-out CWRU bearing-fault loads (92.27 percent on the fourth) but collapsed to 40.64 percent on the JNU benchmark when 1000 rpm was held out, according to an arXiv paper (arXiv:2610.06880v1). The authors report that predictive entropy I1-hat tracked error beyond top-class evidence T-hat and best-competitor evidence F-hat, while decision disagreement I2-hat contributed little, and that standalone Logistic Regression confidence (57.91 percent accuracy on JNU) outperformed the full decomposition. Fusing time-domain and frequency-domain models via Jensen-Shannon divergence beat the model's own entropy (AURC 0.29 vs. 0.36 on the standard split; 0.54 vs. 0.73 under a harder single-condition reproduction), and envelope-spectrum demodulation of literature-verified fault frequencies separated most CWRU fault classes at 99.57 percent using three interpretable features; code, logs and figures were released.

arXiv:2610.06880v1 Announce Type: new Abstract: Machine learning classifiers for bearing fault detection produce scalar confidence scores that conflate confident errors with genuinely ambiguous predictions, and the conventional truth/falsity pair F = 1 - T is algebraically redundant by construction. We operationalize a refined neutrosophic decomposition of a Random Forest + XGBoost + Logistic Regression ensemble into four indicators -- T-hat top-class evidence , F-hat best-competitor evidence , predictive entropy I1-hat, and decision disagreement I2-hat -- evaluated on two bearing benchmarks CWRU and JNU, 600-1000 rpm under a leave-one-condition-out protocol. On CWRU, after correcting a file-to-class mapping error, the ensemble reaches 100.00 percent accuracy on three of four held-out loads 92.27 percent on the fourth , leaving too few errors for uncertainty analysis. On JNU, holding out 1000 rpm, accuracy collapses to 40.64 percent, below a majority-class baseline; Logistic Regression 57.91 percent generalizes far better than the tree ensembles. I1-hat shows a robust association with error beyond T-hat/F-hat, while I2-hat contributes little; standalone Logistic Regression confidence outperforms the full decomposition, a boundary condition we report honestly. Two further results extend this: fusing a time-domain and a frequency-domain model of the same signal and scoring their Jensen-Shannon divergence beats that model own entropy AURC 0.29 vs. 0.36 on the standard split; 0.54 vs. 0.73 under a harder single-condition reproduction , the only indicator moving correctly under a CWRU-versus-JNU distributional-shift contrast; and, on CWRU alone, literature-verified bearing fault frequencies, correctly demodulated via the envelope spectrum, separate most fault classes almost perfectly 99.57 percent using three interpretable features. Code, logs, and figures are released for independent verification.