Predicting the black hole mass and correlations in X-ray reverberating AGNs using neural networks

We develop neural network models to predict the black hole mass using 22 reverberating AGN samples in XMM-Newton archive. The model features include fractional excess variance ($F_{\rm var}$) 2-10 keV band, Fe-K lag amplitude, photon counts and redshift. find that prediction accuracy of is significantly higher than what obtained from traditional linear regression method. Our predicted can be confined within $\pm (2$-5) per cent true value, suggesting technique a promising independent way constrain mass. also apply 21 non-reverberating rule out their possibility exhibit lags (some have too small $F_{\rm var}$, while some large var}$ contradict var}$-lag-mass relation AGN). simulate 3200 multi-feature parameter space investigate global relations if number increases. var}$-mass anti-correlation likely stronger with increasing newly-discovered AGN. Contrarily, maintain lag-mass scaling relation, tight between must preserve. In an extreme case, correlation coefficient decrease and, observed, may suggest extended corona framework where observed are more driven by coronal property rather geometry.