<p>An expeditious growth in the increasing number of Autism Spectrum Disorder (ASD) cases in children has created a precarious situation in the present era. Artificial Intelligence (AI)-based with brain imaging approach can assist doctors in its early detection. An abundant amount of research works have been performed so far using Functional Magnetic Resonance Imaging (fMRI) to detect ASD with an AI-based approaches compared to Structural Magnetic Resonance Imaging (sMRI)-based work. To avoid the large parameter-size based models, this work aims to propose the approach of detecting ASD using a sMRI scan with light-weighted parameter-based classification scheme. This work has proposed a classification scheme which is the hybrid combination of EfficientNet B0 network with the ensemble of Random Forest, K-Nearest Neighbor (KNN), and Extra Trees classifiers. Extensive analysis of the proposed approach has been done on different scales such as data split ratio, number of trees in Random Forest and Extra Trees, number of neighbors in KNN. Also, the comparative analyses with other light-weighted models such as MobileNet, MobileNetV2, and NasNet have been included in the experimentation for presenting the detailed picture. The proposed hybrid classification scheme of EfficientNet-B0 and ensemble of mentioned Machine Learning (ML) classifiers have resulted an accuracy of 86.03%, 87.81%, and 89.56% with the data split ratio of training and testing samples with 70:30, 80:20, and 90:10 respectively. The superiority of the proposed work has been witnessed with the experimental results compared to the sMRI-based state-of-the-art works in detecting ASD.</p>