Life cycle inventory (LCI) data gaps present a crucial challenge in research and development using life cycle assessment (LCA). The present study addresses this challenge by using optimized Artificial Neural Networks (ANNs). ANN's adjustable parameters (i.e. topology and hyperparameters) significantly affect prediction capability. Thus, the optimality of those parameters is of paramount importance. However, performing an optimization on the entire ANN's adjustable parameters is computationally demanding, which may result in an intractable problem. To tackle this challenge, the present study proposes a hybrid approach using heuristics and Genetic Algorithm to systematically design optimal ANNs in a tractable fashion. The proposed approach is then assessed through the prediction of data gaps in the Canadian fuel LCI database, GHGenius (an open-access tool for modeling Canadian fuel pathways). This is achieved using an automation workflow that extracts LCI data from GHGenius at the fuel/province/unit process level. It is found that the resulting optimal ANNs not only are accurate in predicting data to fill CO₂-eq emissions data gaps for unit processes present in the Canadian fuel life cycles but also possess shallower hidden topologies. These salient results are ascribed to the impacts of the input layer on the optimal network. In particular, input layers comprised of categorical and numerical features lead to enhancement of network prediction and/or shallower hidden topology for optimal ANNs. Taken altogether, the present study proposes, implements, and validates a systematic framework to tractably design optimal ANNs for predicting data to fill data gaps in LCI databases. In addition, this study quantifies average contributions of each unit process present in the Canadian fuel life cycles, including fossil-based and renewable pathways, to total CO₂-eq emissions, which is of particular interest in cut-off analysis.