篇名 | Anomaly Detection in Climate Data Using Stacked and Densely Connected Long Short-Term Memory Model |
---|---|
卷期 | 31:4 |
作者 | Teguh Wahyono 、 Yaya Heryadi 、 Haryono Soeparno 、 Bahtiar Saleh Abbas |
頁次 | 042-053 |
關鍵字 | anomaly detection 、 deep learning 、 Denseley Connected 、 Stacked-DC LSTM 、 EI 、 MEDLINE 、 Scopus |
出刊日期 | 202008 |
DOI | 10.3966/199115992020083104004 |
Climate anomalies are considered as an important factor closely related to many disasters causing many human losses, such as airline crash, wildfires, drought and flooding in many areas. Many researchers have projected that the rising global temperature will increase the draught, especially in the mid-latitude areas. Taking those problems into account, studies on anomaly detection in climate are crucial. While climate prediction aims to analyze and model regular pattern of climate, climate anomaly studies aim to model climate deviation from its previous general patterns. Long Short Term Memory (LSTM) is a method employed in this research because it has been proven to work effectively in several anomaly detection studies, especially for data with similar characteristics. This paper presents empirical results of using Basic LSTM, Densely Connected (DC) LSTM, and Stacked-DC LSTM models to detect temperature anomaly as a manifestation of climate change in Semarang City. The results show that Stacked DC LSTM produced higher accuracy in detecting anomaly than the other two methods