Performance evaluation of TinyML and CNN Models on Edge Devices for Real Time Medical Diagnosis

As the use of wearable health devices like smartwatches and fitness bands increases, the need for quick analysis also increases. In these situations, researchers are introducing some new TinyML systems that help to process data directly without the cloud. This study is primarily about the analysis that compares convolutional neural networks (CNNs) and TinyML systems for two medical datasets: diabetes prediction and electrocardiogram (ECG) analysis. In this study, we can clearly see how a model’s complexity is affected to make accurate predictions, the training time, and the quantity of resources. This study shows that TinyML models perform better than other models with less computing power. The TinyML model achieves an accuracy of 85.71% on the diabetes dataset, whereas the CNN model achieves only 76.60%. The TinyML model is more effective when compared with its training time. For the ECG datasets, both models achieve a similar accuracy level of around 71%. The TinyML model trains 1900 times faster than the CNN. This shows that the TinyML model can deliver output much faster than CNN, maintaining the same level of accuracy. From these findings, we can conclude that the lightweight machines give more efficient results. This model enables data analysis without cloud systems. This helps the health monitoring systems to be more practical, affordable, and accessible—especially in remote areas.