Infant crying is a vital communication tool reflecting physiological and emotional states, especially for preterm newborns. Manual analysis of cry signals is subjective and limited in accuracy, necessitating the development of automated systems. This study introduces DCRNet, a novel Deep Convolutional Recurrent Neural Network designed for classifying preterm infant cry signals into five categories: "eair," "neh," "eh," "heh," and "owh." The cry signals of the subjected preterm infant babies in the data collection step in which a longer acquisition period is required. An integrated feature fusion matrix is used for the categorization phase to separate pathological crying created using the integrated features significant to multi-class frequency features retrieved by the Cepstral Coefficients with Bark-Frequency (BFCC), Mel-Frequency (MFCC), and Linear Prediction (LPCC) characteristics. Depending on these features, the sounds of the preterm infant cry are categorised using the DCR Net. The target cry signal has five distinct groups: "owh" for tiredness, "heh" for discomfort, "eh" for burping, "eair" for cramping, and "neh" for hunger. The effectiveness of the proposed DCRNet was evaluated using F1 score, accuracy, precision, recall, and specificity. The proposed DCRNet model achieved a classification accuracy of 97.27%, outperforming state-of-the-art models, including AlexNet, ResNet-101, and VGG-19 enhance the total accuracy value by 7.92%, 6.17%, 4.83%, and 3.49%, respectively.