| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊管理學系 | zh_TW |
| DC.creator | 陸怡廷 | zh_TW |
| DC.creator | Yi-Ting Lu | en_US |
| dc.date.accessioned | 2019-7-16T07:39:07Z | |
| dc.date.available | 2019-7-16T07:39:07Z | |
| dc.date.issued | 2019 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:444/thesis/view_etd.asp?URN=106423032 | |
| dc.contributor.department | 資訊管理學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 在機器學習的過程中為正確反應出資料的全貌，使用大量的數據是不可或缺的，其代價即為運算資源的消耗與浪費，數據中的雜訊或極端值更會影響輸出結果的精確度。本研究提出一個多目標預測模型，透過計算特徵的資訊熵（Entropy）評估該特徵當中所涵蓋的資訊量，從原始數據中挑選出真正對預測目標產生有效貢獻的特徵作為機器學習的輸入值，藉此達到特徵挑選（Feature selection）的主要目的。除此之外，本研究採用球型複數類神經模糊系統（Sphere complex neuro-fuzzy system, SCNFS）搭配由鯨群演算法（Whale Optimization Algorithm, WOA）與遞迴最小平方估計法（Recursive least square estimator, RLSE）所構成的WOA-RLSE複合式學習演算法進行參數訓練，相較於自適應類神經模糊推論系統，SCNFS透過球型複數模糊集合乘載更多維度的資訊量，為多目標預測的主要核心；複合式學習演算法則是藉由將參數進行拆解之後分別以WOA與RLSE訓練之，使模型能夠更有效率地接近最佳解。綜合上述架構所組建而成的預測模型得以吸收多樣化的資料集並且同時預測兩個以上的時間序列數據，最後本文將此模型進一步應用於中國股市的預測，藉由多國金融資料預測多目標股票市場證實中國與他國股市的連動性。 | zh_TW |
| dc.description.abstract | In the process of machine learning, in order to reflect the full picture of the data correctly, the use of a large amount of data is indispensable. Its cost is the consumption and waste of computing resources, the noise or extreme values in the data will affect the accuracy of the output. This study proposes a multi-target prediction model, which estimates the amount of information held by the feature by calculating the information entropy of the feature, and selects the feature that truly contributes to the predicted target from the original data as the input value for machine learning to achieve the main purpose of feature selection. In addition, this study uses the sphere complex neuro-fuzzy system(SCNFS) with a hybrid WOA-RLSE learning algorithm consist of the Whale optimization algorithm and recursive least square estimator, in the architecture of the training model. Compared with the adaptive neuro-fuzzy inference system, SCNFS multiplies the information volume of more dimensions through the sphere complex fuzzy set, which is the main core of multi-target prediction. The hybrid learning algorithm is based on the disassembled parameters, which is trained by WOA and RLSE respectively so that the model can approach the optimal solution more efficiently. The prediction model based on the above architecture can absorb diverse data sets and predict more than two time series data at the same time. Finally, this paper further applies this model to the forecasting of the Chinese stock market, to confirm the relationship between China and the international market by collecting international financial data and multi-target stock market prediction. | en_US |
| DC.subject | 特徵選取 | zh_TW |
| DC.subject | 球型複數模糊集 | zh_TW |
| DC.subject | 球型複數模糊類神經系統 | zh_TW |
| DC.subject | 鯨群演算法 | zh_TW |
| DC.subject | 遞迴最小平方演算法 | zh_TW |
| DC.subject | 多目標預測 | zh_TW |
| DC.subject | Feature selection | en_US |
| DC.subject | Sphere complex fuzzy set | en_US |
| DC.subject | Sphere complex neuro-fuzzy system | en_US |
| DC.subject | Whale optimization algorithm | en_US |
| DC.subject | Recursive least square estimator | en_US |
| DC.subject | Multi-target prediction | en_US |
| DC.title | 特徵選取對智慧型時間序列預測之效能研究 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Research on the Effectiveness of Feature Selection for Intelligent Time Series Prediction | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |