PcaSta‐ASDNet: Principal Component Analysis‐Based Stacking Fusion for fMRI Classification of Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder that manifests during childhood and has a lasting impact on individuals' social interactions and communication. In recent years, research efforts have increasingly focused on leveraging functional magnetic resonance imaging (fMRI) for the classification of individuals with ASD. However, an ongoing challenge is that existing models focus either on effectively distinguishing healthy data from Non‐Healthy data types, or on effectively differentiating between the two subtypes within autism spectrum disorder (ASD)—Autism and Asperger syndrome. For patients, knowing not only that they have autism but also which category of autism they fall under can help them receive timely and effective treatment. To address this problem, our research aims to design a model capable of classification in both scenarios. We firstly designed the Distribution Feature Modular Mean algorithm for determining the dimensionality reduction needed for feature vectors to remove useless information from the feature vectors. Secondly, we adapted the Digit Logarithmic for Reduced Difference algorithm to determine the overfitting and noise cleanup dimensions to address the data imbalance and noise problems. Then, we combined Random Forest and Extreme Gradient Boosting (XGBoost) to build StackingClassifier. We fused the above 2 modules and Stacking Classifier to propose a new stacking integration Algorithm for Autism Spectrum Disorder (PcaSta‐ASDNet). Our proposed model achieves a classification accuracy of 91.78% (Control vs. Non‐Healthy) and 92.77% (Autism vs. Asperger) on the ABIDE I dataset respectively. Also, this model got another classification accuracy of 94.57% (Control vs. Non‐Healthy) and 93.59% (Autism vs. Asperger) on the ABIDE II dataset. The experimental results above show that this method outperforms some existing state‐of‐the‐art classification methods and significantly enhances the differentiation.