<![CDATA[Ticker symbol identification based on stock price data in investor decisions has been proven to be pivotal. Though research exists on stock price forecasting, ticker symbol identification is still a research opportunity. Meanwhile, some temporal-sequential classification methods are available, such as classification-integrated moving average (CIMA) and recurrent neural network (RNN)-based deep learning such as long short-term memory (LSTM), and gated recurrent unit (GRU). Our research aim is to prove that CIMA can perform ticker symbol identification on non-stationary stock price datasets. This research collects ten most well-known stock price dataset from Kaggle and performs pre-processing. Then it designs CIMA with non-stationary data and the benchmark deep learning methods. Both methods are optimized with hyperparameter tuning and model selection between adaptive boosting (AdaBoost) and legacy k-nearest neighbors (KNN). The test results show five non-stationary features in the stock price dataset must go through a differentiation process. Then, AdaBoost has an accuracy of 0.9967 ± 0.001, while KNN has an accuracy of 0.9971 ± 0.001, with no significant difference based on t-test. Meanwhile, AdaBoost has a significantly smaller model size and testing and prediction time than KNN. In benchmarking, CIMA+AdaBoost is superior to the three other methods for accuracy, precision, recall, and f1-score, all of which have a value of 0.996. Our research contribution is ticker symbol identification based on stock price using CIMA on multiple-class sequential classification with non-stationary data. For future research, we advice to perform this method on other stock price data.]]>
Copyrights © 2024
