Stock price prediction is a popular research field in the financial and big data industries. Over decades, researchers have leveraged various machine learning and deep learning algorithms to build many innovative models to predict stock prices, and improve the performance a lot. Neural network algorithms are the most choice for recent researchers, and constructing hybrid models with more than one algorithm is demonstrated to have a better performance. However, the machine learning algorithm EXtreme Gradient Boosting (XGBoost) also has its unique advantages compared to other algorithms, and few researchers use it to conduct stock price prediction experiments. Therefore, this thesis proposes a creative method to construct prediction models based on the XGBoost algorithm. The results show that among all 14 XGBoost models, 13 models have excellent performance, with the MAE/MINₛtockₚrice values lower than 4% as well as R2 more than 0. 99. Only one model based on IBM dataset just reach to an acceptable performance. Overall, the XGBoost algorithm has high feasibility and generalization ability in the stock price prediction field. This thesis will inspire future researchers to consider the XGBoost algorithm to build hybrid models and advance the prediction performance.
Songbo Zhai (Mon,) studied this question.