Efficient data processing method for edge intelligence based on SVM

Recently, edge intelligence is emerging. It is widely used because it allows you to make business models more efficient through edge intelligence, but there is still a lack of models to use for edge intelligence. Artificial neural networks are being used a lot because they solve the problem of overfitting, which is a disadvantage of artificial neural networks and increase the speed by improving GPU performance. However as the independent variable to be learned increases, the learning time and characteristics of the artificial neural network slow down and the size of the neural network becomes very heavier. The shortage of these artificial neural networks make it difficult to make edge intelligence for classification or prediction in real life. Manufacturing lightweight models using heavy neural networks and high-performance GPUs also presents a lot of financial problems. However, SVM based on statistical mathematics exists between machine learning. Although it is relatively inferior to artificial intelligence, it is better light and has stable performing. To do this, we propose an efficient data processing method that improves the characteristics of the SVM (the overall performance, accuracy and learning speed of the SVM) and makes it easier for the SVM to learn the data. When four types of data processing were applied, the accuracy increased by more than 0.4% for the iris data set and 2% on average for the other data sets, and the training time for all data sets was also reduced. As a result, SVMs that can process data efficiently are much lighter than artificial neural networks and have very good training times versus accuracy. Therefore, in this paper, we propose an efficient data processing method based on SVM for edge intelligence.