時間序列資料的變化有著眾多變因，在預測上一直是具有挑戰性的問題和研究。最常應用於股市上的股價變化，從時間的推移中找出股票之間的關係，因此本篇設計一多目標時間序列預測模型，應用於股價預測。模型包含兩種模型架構，人工神經網路（Artificial nural networks; ANN）及球型複數神經模糊系統（Sphere complex neuro-fuzzy system; SCNFS）。在SCNFS的部分，加入球形複數模糊集（Sphere complex fuzzy sets; SCFS）及箭靶層（Aim object layer），前者產生之歸屬程度為複數空間上之單位球體，使模型可進行多目標運算；後者以多對多的因果關係，降低模型後鑑部的運算負荷。在資料前處理的部分，加入多目標特徵挑選，以亂度熵（Entropy）的概念，從龐大的資料集中挑選出對目標具貢獻的輸入資料。在機器學習的部分，使用混合式機器學習演算法，結合連續型蟻群演算法（Continuous ant colony optimization; CACO）及遞迴最小平方估計法（Recursive least squares estimation; RLSE），有效地訓練模型的參數。本篇共執行三個實驗，實驗一透過單目標預測驗證ANN-SCNFS的效能及CACO-RLSE的優化效果；實驗二以相同資料集同時進行單目標預測及多目標預測，驗證模型可一次預測多目標之能力；實驗三透過不同標的股進行多目標預測，研究資料集在使用上的彈性。實驗結果表明本研究提出之預測結果大多優於文獻結果，顯示在股票的時間序列預測上有良好的效能。

Purpose － Time series forecasting is a challenging research issue. In the past research, most of them were single-target forecasting. However, in the stock market, stocks will effect each other. Therefore, we hope to predict the stock price on multiple targets simultaneously, and find the relationship between them through this research. Design/methodology/approach－In the data preprocessing phase, we use multi-target feature selection to find the useful data for prediction. Then, in order to predict multiple targets, the proposed model in the paper combines artificial neural networks (ANN) and asymmetric sphere complex neuro-fuzzy system (SCNFS). Moreover, we use the hybrid machine learning algorithm CACO-RLSE, which combines continuous ant colony optimization algorithm (CACO) and recursive least squares estimation (RLSE) to train the parameters in the model. Findings－The result shows that the performance of multi-target prediction is better than single-target prediction. In addition, the use of input dataset is flexible, we can use US index to predict US stock price. Research limitations/implications－This study only focused on US stock market information. In the future, we plan to research the different stock market, and get more information to do multi-target time-series forecasting. Practical implications－In this paper, we have proposed a multi-target time-series forecasting model. This method can give the investors some support to make investment decisions. Originality/value－We use hybrid machine learning algorithm, CACO-RLSE, to improve the efficiency of training parameters. Besides that, we change the structure of original SCNFS, and let it become an asymmetric model structure by using the novel aim-object layer (AOL) in the model.