The Comparison of Some Version of Linear Vector Quantization (LVQ) for Vitamin and Mineral Deficiency Early Detection

Abstract

Vitamin and mineral deficiency are often ignored because they do not have a direct impact on body health. However, prolonged deficiency can cause various diseases from mild to serious illness. Some previous research in computer science already conducted to make early detection of vitamin and mineral deficiency, but no one has produced an adaptive model to find out the most dominant type of deficiency. Therefore, the goal of this research is to develop an adaptive model using an artificial neural network (ANN) with Linear Vector Quantization (LVQ) as the learning algorithm to make early detection of vitamin and mineral deficiency. LVQ consists of three layers: an input layer that represents the features, output layer that represent the class label, and the competitive layer. The competitive layer will save the distance between the input vector and the codebook vector from each class. The distance will calculate using Euclidean Distance. LVQ also involves some parameters in the training process, like epsilon value, learning rate, codebook vector, epoch, and window size which obtained by trial and error experiment. This research will also compare the performance of some version of LVQ. The experiment results show that the maximum accuracy level obtained by the system is 85.71% by using LVQ3. The dataset used split into data training and data testing with a ratio 84:16 respectively. From our scenario, the optimum model was achieved by using 20 codebook vectors with the number of epochs is 3400 and the value of the learning rate parameter (&agr;) of 0.4, window size (ō) of 0.3, and epsilon (ε) of 0.2.