Environmental factors such as climate change, pollution, and adverse weather directly impact human health, driving many households to maintain a mini medicine cabinet. However, correct drug usage requires medical knowledge, and mislabeling can lead to misuse. We propose a Decision Support System (DSS) that (i) applies Natural Language Processing (NLP) to extract symptom keywords, (ii) uses a Deep Neural Network (DNN) to predict diseases, and (iii) leverages an optimized Convolutional Neural Network (CNN) to recognize and classify medication images. In evaluation on a synthetic disease–symptom dataset (246 000 samples, 773 diseases), our DNN achieved 86.74% test accuracy (loss 0.4793), outperforming classical classifiers such as Random Forest (84.23%) and Naive Bayes (83.65%) under default settings. For drug images, our MobileNetV2-based CNN reached 98.05% accuracy (loss 0.2196), significantly surpassing default MobileNetV2 (92.70%) and other benchmark CNNs (85%). A small–scale test on 10 unseen pill images yielded a perfect 9/10 recognition rate. These results demonstrate that our lightweight, end–to–end DSS can both improve diagnostic accuracy over existing approaches and run efficiently on mobile or web platforms.