Neural STATIS: A Nonlinear Multiblock Consensus Framework via Learned Latent Representations

Multivariate analysis of multi-table databases presents significant challenges when underlying structures exhibit nonlinear and time-varying behavior, limiting the applicability of classical linear methods such as STATIS. This study introduces Neural STATIS, a novel methodological approach that integrates the representational capacity of neural networks with the structural logic of classical STATIS to capture nonlinear dependencies across multiple data tables. Longitudinal OECD data encompassing socioeconomic and sustainable development variables across multiple time intervals were used to evaluate the method’s feasibility and enable direct comparison with traditional STATIS. Neural STATIS successfully identified complex trajectories and subtle temporal patterns that classical linear analysis failed to detect, providing deeper visualization of dynamic relationships across tables. These findings demonstrate that Neural STATIS offers a more adaptive and comprehensive analytical framework than conventional linear methods, particularly in contexts characterized by ambiguity and nonlinearity. Grounded in Neutrosophic philosophy, which explicitly acknowledges uncertainty as an inherent component of complex phenomena, this approach broadens the scope of multi-table analysis and facilitates more informed decision-making in real-world settings where linearity assumptions are not sustainable.