PENGENALAN AKTIVITAS MANUSIA DENGAN MENGGUNAKAN METODE CONVOLUTIONAL NEURAL NETWORK (CNN)

The development of the times that is increasing rapidly makes the current generation or what is called the millennial generation become familiar with several developing technologies, including sensor technology. The commonly used sensor is the Accelerometer. Every day humans are always physically active, recognizing human activity or what is commonly called Human Activity Recognition (HAR). In this study using the WISDM dataset provided by the Wireless Sensor Data Mining (WISDM) Lab. The addition of the dropout layer in this study has been successfully implemented with an overall average accuracy above 96.26% and a loss of 0.14%. Based on the test results that have been carried out on the CNN method, it can be concluded that the test of classification evaluation results using the Convolutional Neural Network (CNN) method has been calculated based on the average. For the All Activity Confuison Matrix section (96.26%), the Static Activity Confusion Matrix section (97.22%), the Dynamic Activity Confusin Matrix section (97.18%). Compared to previous research that also uses a dropout layer, the accuracy in this study has increased for Downstair, Sitting and Upstair activities. Jogging and Standing activities decreased by 1% and Walking activities decreased by 4%.