Assessment of Scientific Creative-Potential by Near-Infrared Spectroscopy Using Brain-Network-Based Deep-Fuzzy Classifier

The work presents a novel approach to assess the scientific creative ability of subjects by analyzing their brain connectivity patterns through functional Near-Infrared Spectroscopy (fNIRS) during participation in an analogical reasoning test. The proposed method involves three key stages: i) construction of brain connectivity networks using Wavelet Transform Coherence (WTC), ii) abstraction and analysis of three node-based network features, and iii) classification of abstracted features into five degrees of creative potential by a novel Enhanced Graph Convolution Induced Type-2 Fuzzy Classifier (EGCIFC). The novelty of the classifier lies in: i) design of an enhanced graph convolution operation that encapsulates local and global structural information from the input graph, ii) use of the Smish activation function to improve performance, iii) inclusion of a one-dimensional spatial convolution layer for preserving relevant information within convolved embeddings, iv) design of a novel mapping function to mitigate uncertainty among the spatial convolved vectors in the type-2 fuzzy layer, and v) application of Takagi-Sugeno-Kang (TSK)-based fuzzy reasoning to reduce computational cost. Evaluation on three datasets, each comprising over 45 individuals from different scientific backgrounds, shows that EGCIFC improves classification accuracy by 2.25% over the nearest competitor and by 22.72% over the lowest-performing baseline. The proposed method also reduces computational cost by 7.46% and 54.7% compared to the nearest and worst competitors, respectively. Additionally, EGCIFC exhibits a standard deviation of ±0.72% in classification accuracy, reflecting its robustness. Hence, the proposed approach may prove effective for recruiting individuals with varying degrees of scientific creativity across different research sectors.