Owing to the day-to-day developments in the modern environment and food nature, humans are affected by various types of diseases like Covid-19. The disease recovering stage needs prolonged treatment, which demands various drugs based on the patient's health condition. The main problem is the drug recommendation and the drug compound molecules that are not relational to curing diseases that lead to side effects. So, knowledge analyzing factors based on the relational feature selection approach are needed to make the best recommendation. To improve the performance of drug recommendations based on a machine learning approach, a Spectral Deep Feature Classification based Compound Analyse Model (SDFC-CAM) is intended for analyzing the success rate of drug recommendations. The features are processed with Mutual Spectral Scaling Feature Selection (MS2FS) filter and inputs are given to Inter-Segment Sigmoid Activation Adaptive Recurrent Neural Network (InS2ARNN) to make recommended classes. The system initially pre-processes all the patients' nature and their recommended drug molecule compounds as a collective dataset. The feature selection is carried out to generate relational patterns using the distance vector features correlated to mutual molecules weights. The features are selected from the success rate influence measure generated by the candidate selection model (CSM). The estimated rate of features weights that are ruled to create the activation function is called the intra-class activation function. Further, the activation fixes the logical rules to make predictions fed to the feature weights into the neural classifier. The RNN predicts the results which are trained based on logical rules iterated in the neural classifier. The classifier makes successive predictions that are grown up to make the right recommendation of compound molecules to reduce the risk of the side effects. The proposed SDFC-CAM with InS2ARNN classification algorithm aims to provide efficient drug recommendations. The proposed algorithm achieves 94% of prediction results for drug recommendations compared to the existing algorithms.