This paper investigates deep learning algorithms to aid the evaluation of sustainable aviation fuels (SAF) using comprehensive two-dimensional gas chromatography (GCxGC). Specifically, two tasks are addressed: 1) Detection of low-confidence fuels with novel chemical characteristics from reference fuels; this is achieved via an autoencoding neural network to reductively encode all available GCxGC information of target fuels to latent representations of highest variance and optimal clustering properties to compute an overall novelty score per fuel. 2) On “normal” fuels, the feasibility of learning-based predictions of key physicochemical properties such as density, distillation, flash-point, and kinematic viscosity from the aforementioned GCxGC features using artificial neural networks. These tasks are demonstrated on a highly diverse jet fuel dataset comprised of 106 samples including petroleum-based aviation fuels and SAF derived from various sources, methods, and blending ratios. The GCxGC data covers over 80 hydrocarbon groups within aromatics, iso-paraffins, n-paraffins, and cy-cloparaffins of various carbon numbers. Results show that the proposed novelty detection scheme is successful in preemptively identifying chemically novel fuels that exhibit high predictive errors when evaluated on downstream models. Furthermore, the neural network-based fuel property predictions were found to be superior to the traditional, linear partial least squares regression model, despite the relatively small dataset size and the large number of GCxGC features utilized, both of which are known to compromise modelling of large neural networks.