DeFN:	The standard forecasteing model. The advantage is that it misses very few flares. It can predict flares that will occur within 24 hours for "X-class", "M-class and above", and "C-class and above". (Nishizuka et al. 2018 ApJ) [Download here]
	DeFN-Reliable (DeFN-R):	The model with improved reliability of DeFN. The predicted probability presents the actual occurrence frequency. The user can set their own threshold for prediction. (Nishizuka et al. 2020 ApJ) [Here]
	DeFN-Quadro (DeFN-Q):	The model that extends DeFN to 4-class prediction of "X, M, C, None". Flare size can be known more accurately. Coronal Mass Ejection (CME) prediction model is also added. (→ CME NET) (Nishizuka et al. submitted.)
	Operational DeFN:	An extention of DeFN to a real-time operational system. (Nishizuka et al. 2021 EPS) [Here]

Other References:

About AR detection method and feature extractions in Deep Flare Net
- (1) Solar Flare Prediction Model with Three Machine-learning Algorithmjs using Ultraviolet Brightening and Vector Magnetograms
      (Nishizuka et al. 2017 ApJ) [Here]
      Active region detection methods, feature extractions and feature rankings are discussed in this paper.