以機器學習為基礎之股價漲跌預測模型-以電子股為例

隨著疫情時代的來臨，所衍生出的消費性需求、物價通膨、無限QE等各項政策，帶動著全球股市的另一新動能，也包括台灣股市，從疫情開始的2020年的8,523點，一路創高至2022年的18,619，掀起了台股的另一波牛市，因此在股市預測各個議題當中，也讓資訊、財經、資料科學等領域的專業人員積極投入在這議題當中。本研究此次針對晶圓代工第二的聯電進行機器學習分析，所使用機器學習方法為支援向量機、隨機森林、神經網絡等三項機器學習工具，針對3大構面、50個指標進行預測，結果顯示，丟入50個指標進行預測，隨機森林準確率85.49%、支援向量機準確率85.04%、神經網絡準確率82.85%，從中可以看出隨機森林的機器學習模型，做預測時，擁有較高的準確率，此外本研究也發現到，在股市中因各種環節環環相扣，因此會受到各種不同的因子干擾其預測結果，在隨林森林的模型中，可以看出，在做因子縮減至10個指標時，準確率達到85.94%，相較於另外兩種機器學習模型，擁有較好的預測效率。

關鍵字

股市預測；隨機森林；支援向量機；神經網絡

並列摘要

With the advent of the COVID-19 pandemic era, consumer demand, price inflation, unconstrained QE and other governmental policies have rallied stock markets around the world. Taiwan is no exception; from 8,523 points in 2020, when the epidemic began, it has since reached a record high of 18,619 points in 2022, setting off another bull market for Taiwan’s TWSE. Therefore, professionals in Taiwan’s tech, finance and data science are actively involved in various issues of stock market forecasting. In this study, machine learning analysis is conducted for UMC, Taiwan’s second largest semiconductor manufacturer. Three machine learning tools are used to predict 3 major dimensions and 50 factors, namely support vector machines (SVM), random forests and neural networks. The results indicated that the accuracy of random forest was 85.49%, the accuracy of SVM was 85.04%, and the accuracy of neural network was 82.85 %when 50 factors were used for predictions. From this, we can observe that the random forest method has a higher accuracy rate when making predictions. In addition, this study also found that the various components of the stock market are interrelated and therefore subject to various factors that can interfere with the prediction results. In the random forest model, we can see that the accuracy rate reached 85.94%hen factors are reduced to 10, which presents a better prediction efficiency compared to the other two machine learning models.