This thesis proposes methods employing an evolving Spiking Neural Network (SNN) architecture for the analysis of spatio-temporal neuroimaging data. Multivariate Spatio-Temporal Brain Data (STBD) is intrinsically complex as it contains both time and space dimensions that represent the patterns of cognitive processes in the brain. Scrutinising the spatio-temporal interactions between variables in such complex data demands incorporating the spatial and temporal aspects into the model’s computations. To this end, first an SNN architecture was used for modelling, learning, mapping and classifying of STBD, including Electroencephalogram (EEG) and Functional Magnetic Resonance Imaging (fMRI) data. I designed SNN models that allowed for a better understanding of cognitive processes by capturing the spatio-temporal interactions between variables when compared with extant reservoir computing systems. The models enhanced the classification performance by achieving up to 92% accuracy which represents an average improvement of 20% when compared with different machine learning methods. Further, I proposed and developed a new dynamic spatio-temporal clustering approach which allowed for the assessment of the evolving learning patterns in SNN models. This study led to knowledge discovery in SNN evolutionary learning patterns and resulted in feature selection that improved the classification accuracy by up to 10%. It also revealed the trajectory of brain areas involved in response to a cognitive task. The proposed clustering configuration was evaluated using a validity measurement method based on cohesion and separation that represented a high goodness of the clustering structure. Finally, I proposed a new personalised modelling approach for integrated static and spatio-temporal data using SNN models. To build a personalised SNN model (PSNN), I developed a new clustering method, named Dynamic Weighted-Weighted Distance K-nearest Neighbours (DWWKNN). The developed PSNN improved the classification accuracy by 12% when compared with the global SNN models. This also resulted in creating a profile for an individual. Overall, this research has scrutinised the hidden evolutionary learning patterns in SNN architecture, which resulted in an identification of neural areas activated by different input neurons. Furthermore, it has demonstrated an original personalised modelling that resulted in an improvement in classification accuracy.