The effectiveness of Random Forest (RF) and Convolutional Neural Network (CNN) models for dividing lung sounds into four groups asthma, COPD, healthy and pneumonia is compared in this study. CNNs were trained on spectrogram images to capture intricate sound patterns, while RF, a conventional model, was assessed for its effectiveness on structured variables such as Mel Frequency Cepstral Coefficients. The RF model outperformed the CNN model in terms of generalisation, even though it was able to capture intricate patterns. This was particularly true for the COPD and Healthy classes, according to the results. In order to balance accuracy and efficiency, RF is advised for deployment because to the Raspberry Pi’s processing constraints. Suggestions for hybrid models, transfer learning and data augmentation are explored in order to improve performance even further. This work advances the creation of portable, real-time diagnostic devices.