大數據分析技術結合時間序列預測應用－以區間式改良神經網路為例

Application of Big Data Analysis Combined with Time Series Prediction-Using Interval Improved Neural Networks

林秀怡(Hsio-Yi Lin) ； 徐彬偉(Bin-Wei Hsu)

《中正嶺學報》 52卷1期 (2023/05) Pp. 89-100

https://doi.org/10.30188/JCCIT.202305_52(1).0008

摘要

文獻中常利用倒傳遞式類神經網路（BPN）、長短期記憶（LSTM）等深度學習模型進行時序性預測，並以點估計值為預測方法，然而點估計卻常造成預測準確率趨近0之現象。本研究提出一區間式LSTM模型，並與傳統BPN及LSTM進行預測精確度比較。以新冠肺炎爆發後台灣生技指數日收盤指數為標的，結果顯示LSTM之RMSE、MAPE及MAE均優於BPN模型。區間式LSTM分別在信賴水準99%、95%與68%測試樣本指數值的準確預測率分別達100%、99.33%與94%，實可提升深度學習模型於時序性預測之實用性，疫情期間若應用於國軍確診個案數預測，提升疫情防控策略措施之有效性，以提供後續國軍醫療資源超前部署。

關鍵字

長短期記憶 ； 倒傳遞式類神經網路 ； 疫情防控措施 ； 深度學習

並列摘要

In the literatures, deep learning models such as backpropagation neural network (BPN) and long short-term memory (LSTM) are often used for time series prediction, and point estimation is used as prediction method. However, point estimation often causes the phenomenon of prediction accuracy to approach 0. This study proposed an interval-type LSTM model, and compared the prediction accuracy with traditional BPN and LSTM models. Taking the daily closing index of Taiwan's listed biotechnology and medical index after the outbreak of the new crown pneumonia as the target, the results show that the RMSE, MAPE and MAE of LSTM are all better than the BPN model. Interval-based LSTM has an accurate prediction rate of 100%, 99.33% and 94% respectively at the confidence level of 99%, 95% and 68% of the test sample index values, which can really improve the practicability of deep learning models in time series prediction. During the epidemic, if it was applied to the prediction of the number of confirmed cases in the national army, it would improve the effectiveness of the epidemic prevention and control strategies and measures, so as to provide the follow-up medical resources for the advance deployment of the national army.

並列關鍵字

LSTM ； BPN ； epidemic prevention and control strategies and measures ； deep learning