An Improved Multi-Scale Adaptive Deep Deterministic Policy Gradient-Based Behavior Pattern Analytics Model for Learning and Action in Children
E. Samatha Sree Chaturved, L. Mary Gladence · International Journal of Human-Computer Interaction · 2025
This paper presents novel behavior pattern analytics approach for action recommendation and learning in children. At first, from the standard data sources, the multimodal data, including Electroencephalogram (EEG) signals, speech signals, images and language are fetched. After that, the gathered multimodal data are forwarded to the feature extraction phase, where Term Frequency-Inverse Document Frequency (TF-IDF) and Convolution Neural Network (CNN) are used to extract the language and image features, respectively. Moreover, from the speech signal, the spectral and Cepstral features are extracted and then, from the EEG signals, the EEG signal features are extracted. After that, the extracted features are integrated and fed into the learning and action recommendation module, where the Multi-scale Adaptive Deep Deterministic Policy Gradient (MADDPG) is employed. Here, the Enhanced Cheetah Optimizer (ECO) is introduced for optimizing the MADDPG technique’s parameters. At last, an experimental analysis is performed for the recommended model to show its efficiency.