Deep learning-based smart wearable sensors healthcare model for fitness trackers

As smart wearable sensors become more commonplace, fitness trackers offer a wealth of data for getting a full picture of people’s health. Anomaly identification, however, may be necessary since their multidimensional activity data may include unknown anomalies. Because of the “curse of dimensionality,” conventional density estimation methods are notoriously difficult to implement, leading to subpar detection capabilities. This issue was tackled by utilizing a Gaussian Blend Generative Model (GMGM) to remove medical services records, resolving this issue. By using a Variational Autoencoder (VAE) to train the model on raw data, latent properties can be extracted, and the reconstruction error can be reduced. Following this, it creates a Profound Conviction Organization (DBN) conjecture on the blended participation of tests by using the idle conveyance and recovered features. The Gaussian Mixture Model (GMM), DBN, and VAE are then optimized simultaneously to avoid model decoupling. By making thickness expectations for all data of interest the Gaussian Blend Model recognizes exception tests as those with densities higher than the preparation limit. The GMGM’s area under the curve metric was 5.5% higher than that of the Deep Autoencoder GMM (DAGMM) when tested on the ODDS standard dataset. Finally, additional proof of the method’s efficacy was provided by the outcomes of experiments carried out on actual datasets.