<p>Anxiety is a chronic mental disorder triggered by stressful situations potentially leading to serious social and health issues. This research aims to develop a method for automatically identifying various anxiety levels based on electrocardiographic (ECG) signals, as this signal is far cheaper and easier to acquire than an electroencephalogram (EEG) signal. Publicly available ECG data were analyzed, which were collected from 19 subjects using wearable sensors under anxiety-inducing video stimuli. Based on the Hamilton anxiety rating scale, these subjects were categorized into four distinct groups: normal, light, moderate, and severe anxiety. The wavelet scattering network (WSN) is an optimized deep convolutional network designed for feature extraction and the effective revelation of discriminative patterns within signals. Moreover, WSNs are robust to local perturbation, rendering further reliability and performance to the networks. The extracted feature was tested on different machine learning (ML) and deep learning (DL) models to find the best classifier for it. The developed WSN-BiLSTM framework, which includes a WSN for feature extraction and a Bidirectional Long Short-Term Memory (BiLSTM) model for classification, performs exceedingly well and is better than typical machine learning models applied. The proposed WSN-biLSTM model achieves an accuracy of 99.79% and a Kappa score of 99.71% with 10-fold cross-validation, thereby surpassing recent studies by 7.52% in accuracy and by 10.71% in Kappa scores. The developed approach proved to be simple, computationally efficient, with only 470,532 trainable parameters, and gave higher classification accuracy; hence, it is clinically relevant for the diagnosis and management of anxiety disorders.</p>