<p>The application of thermal performance prediction in solar stills (SS) remains of utmost significance in improving the performance of desalination. However, traditional methods have the following main drawbacks: lower accuracy of prediction, a relatively large number of calculations during the model construction, and ineffective feature selection that affects the work’s efficiency. To address these problems, this present research introduces a new Spiking Neural Network (SNN) model that provides feature selection through the Pufferfish Optimization Algorithm (POA) and then utilizes the Chaotic Single Candidate Optimizer (CSCO) for training. The experimental setup involves two identical single-basin, single-slope solar stills: the first having a polycarbonate (PCSS) compound parabolic concentrator and the second an ALSS assembly with an air gap under the absorber plate. The above qualities of aluminum such as thermal conductivity, anticorrosive nature, cost and lightweight make ALSS suitable for solar stills. The solar irradiance, ambient temperature, water temperature, and plate temperature were input data for this study to estimate energy efficiency, exergy efficiency, heat transfer (HT) coefficients, and distillate output (DO). The temporal inconsistencies are resolved by the Enhancing Temporal Consistency (ETC) constraint, and hyperparameters are adjusted by the two-phase SCO strategy in SNN-CSCO model. The analytical results suggest that the proposed model has high accuracy with the adjusted R-squared of 0.99, RMSE of 0.01, low distillate output error of 0.009, and HTC error of 0.31, while has better performance in comparison with the existing methods. The SNN-CSCO strategy is therefore developed and adopted as the best practice for better means of desalination.</p>