The complex and multi-faceted endeavor of ascertaining a child's psychological state involves cognitive, emotional, and behavioral development. Traditional diagnostic techniques have restrictions based on subjectivity and clinician-based competency that could further delay treatment. AI has the potential to transform this process by determining psychological states through data-driven, automated, and scalable classification tasks. This paper proposes a blended ensemble model that combines the Long Short-Term Memory (LSTM) network for modeling sequential data from speech and behaviors, the Convolution Neural Network (CNN) for identifying spatial features from facial imagery data, and Random Forests (RF) for classifying psychological states using decision-level fusion. The assessment utilizes multimodal techniques of FER+ (facial emotions), CHILDES (speech transcripts), and a survey for kid behavioral data. The hybrid ensemble model is called CNN-LSTM-RF from the convolution, recurrent, and decision-making parts of the system.