Children Autism Detection Using Residual CNN and EfficientNetB1

Autism is a developmental disorder that can cause serious social, communication, and behavioral issues. Early intervention for autistic children can aid in improving cognitive ability and reducing autistic symptoms. In this study, we investigate how well Convolutional Neural Network (CNN) models perform when estimating automated engagement from autistic children's faces. The main contribution is to create a CNN residual using residual blocks containing Depthwise convolution layers (DCL) and convolution layers to reduce the number of parameters in the model by treating each channel of the input image separately with a Depthwise kernel. A Residual CNN model and EfficientNetB1 have been proposed for autism image classification. The EfficientNetB1 architecture with Global Average Pooling (GAP) layer is used for scaling dimensions, and a classification layer is done with a SoftMax layer. The Kaggle dataset, which contains 2,940 images of faces, is used in the experiments. The model's performance was evaluated and compared to existing models in terms of precision, recall, f1-score, and overall accuracy. The two models used achieved classification rates of 86.33% and 88%, respectively, for Residual CNN and EfficientNetB1. According to the results, the method outperforms others using the same dataset.