CkLR Algorithm for Improvement of Data Prediction and Accuracy Based on Clustering Data
Se-Hoon Jung, Jong-Chan Kim · International Journal of Software Engineering and Knowledge Engineering · 2019
In this paper, we proposed the CkLR algorithm to build a model to analyze the response variables of clustered outcomes with a K-means algorithm to categorize and analyze unstructured data such as SNS and assess the prediction rate of clustering and its accuracy rate after the entry of new data. CkLR performs a neural network algorithm based on the clustering outcomes of the data classified in the previous stage. The CkLR model applies a neural network to reflect the entire form of data learning methods rather than stepwise data learning based on clustering data, and conducts research to select a neural network model and analyze variables through learning by clustering. The datasets of the input clustering outcomes are adjusted at certain percentages (learning, testing, and evaluation) to check the prediction rate of input outcomes. The study also proposed a CkLR algorithm based on an LSTM recurrent neural network to compensate for the problem of determining the size of input and output value data for each repetition in a single training session. The training structure of the proposed prediction model, based on a neural network, would calculate the loss values of training, minimize them, and select an optimal model reflecting weights in case of clustering sequential data. The performance evaluation results demonstrate that it recorded prediction and accuracy rates that were 3.65% higher than the old single neural network structure through individual neural networks.