Optimization of SV-kNNC using Silhouette Coefficient and LMKNN for Stock Price Prediction

A broker often finds it difficult to decide to buy or sell shares. Due to the large amount of stock data and errors in analyzing stock indicators, it can cause the broker to suffer losses. Predicting stock prices with high accuracy becomes onerous. Although unnecessary variables have been excluded but there are still outliers. It is still difficult to attain the most optimal class value at the data clustering stage. The objective of this research is to propose a model to refine the accuracy of stock prediction and to produce the best class at the data clustering stage. The proposed model is named SV-kNNC - Silhouette Coefficient - SOM (Self Organizing Map) - LMKNN (Local Mean-based K Nearest Neighbor). First, the best class value is obtained using the Silhouette Coefficient, then the data is weighted, finally the stock data is classified using the LMKNN to secure the prediction results. In the testing section, SV-kNNC - SOM + Silhouette Coefficient - LMKNN is compared against SV-kNNC - SOM - KNN (K Nearest Neighbor). Confusion Matrix is used during the test to get the predictive accuracy value. It can be seen from the results of the prediction accuracy of BBCA's shares is 86.99% with the best K is 3, ASII company is 78.30% with the best K is 3, TLKM company is 85.10% with the best K is 5, BBRI company is 84.73% with the best K is 6, PGAS company is 71.36% with the best K is 5.