在景氣認定上，往往需要蒐集足夠的資料、觀察夠長的時間，才得以確認景氣是否發生轉折。因此，發布認定結果的時點往往較實際轉折落後一段頗長的時間，各界難以即時得知當前的景氣狀態。本文參照Giusto and Piger（2017），應用機器學習上的學習式向量量化（Learning Vector Quantization, LVQ）方法，並設定若干假設及判定規則，即時認定台灣2000年以後的景氣循環。LVQ方法毋須對景氣循環的資料生成過程（data generating process）做任何假設，適合台灣在不同經濟發展階段下，景氣循環亦具有不同特性之情形。實證結果發現，藉由LVQ方法，可大幅縮短景氣轉折發生後所需的認定時間，且得到的轉折時點與國發會相近，故應有助在正式發布認定結果前，得到關於當前景氣狀態的參考資訊。

The official chronologies of business cycle turning points often suffer from a substantial time lag, which makes it difficult for economic agents to identify the starting point of a new business cycle phase. Following Giusto and Piger (2017), this paper identifies Taiwan's business cycle turning points after 2000s in real time using a machine learning algorithm known as Learning Vector Quantization (LVQ). Since LVQ does not rely on the specification of the business cycle's data generating process, it is suitable for addressing distinctive features in Taiwan's business cycles at different stages of economic development. Utilizing an LVQ algorithm, business cycle turning points can be identified quickly with a lag of between seven and ten months, considerably better than the official's 12 and 51 months. Furthermore, the empirical results suggest that the turning points identified by LVQ and the official method are quite consistent, and their differences are within a three-month range. In contrast, the turning points estimated by Markov-switching models are significantly different from the official turning points.