Chest X-ray (CXR) imaging can detect and classify COVID-19, pneumonia viral, and normal classes early. Due to poor feature selection, classical AI methods performed poorly in categorization. This study created CIDC-Net for CXR image-based disease categorization using optimal feature analysis. To improve CXR images, adaptive median filter (AMF) was used to reduce noise and preserve structural details. Inspired by the plant's adaptive waterwheel mechanism, modified waterwheel plant optimization (MWWPO) algorithm for feature selection follows. The MWWPO automatically identifies the most useful disease-specific, disease-dependent features from CXR pictures, reducing overfitting and computational complexity and improving classification performance. Finally, an MLCNN is trained on MWWPO characteristics to identify COVID-19 from other respiratory illnesses. Depth and convolutional layers allow the network to learn hierarchical representations and capture detailed patterns and spatial information in CXR images. This study verifies COVID-19 detection and classification results on a large and heterogeneous CXR image collection, showing its high accuracy, sensitivity, and specificity.