台灣的地理環境，讓海上交通運輸非常成熟。同時，也讓海路走私的問題非常嚴重。 行政院於 2002 年建置海巡資訊系統，儲存船隻進出港口紀錄，以及執法人員安全檢查 的結果。由於目前的檢查方式，準確率僅略高於 5%。因此本研究嘗試藉由資料採礦的 方法去改善此一現況。然而，走私的紀錄與許多的資料集相同，均存在不平衡資料的問 題。因為絕大部分的分類器在面臨不平衡資料時，容易產生嚴重的偏誤。因此，本研究 結合決策樹、人工神經網路，以及支援向量機，分別使用 Bagging、Boosting、Stacking 等三種合議法，建構分類集成模型，以改善分類方法對於走私資料的分類效益。同時， 藉由此一過程，探究何種分類集成模型對於走私資料的分類效益最佳。研究結果顯示， 使用決策樹的分類方法效果最佳。本研究的結果能提高分類效能與提升海巡執法機關對 走私船隻之識別率，進而捍衛國境安全。

Due to the location of Taiwan, the sea transportation has become very mature; likewise the sea smuggling has converted into a serious problem. The Coast Guard Information System built in 2002 by the Executive Yuan saves the records of the in-and-out vessels of ports, as well as the results of the safety check by law enforcement officers. For the low accuracy rate of the current checking, slightly higher than 5%, this study attempts to improve the current situation via data mining. Still, the imbalanced data, which leads a serious bias while encountering with most of the classifiers, happen in the records of smuggling as well as many other datasets. This study apply decision trees, artificial neural networks and support vector machine to construct an integrated classification model for the efficiency of classifying the data of smuggling via bagging, boosting and stacking respectively. In the process, the most efficient integrated classification model for the data of smuggling is investigated. The findings suggest that the multi-classifier has performed ever far better than the traditional single classifier. The results of this study can improve the classification efficiency and the recognition rate of the Coast Guard Administration on smuggled vessels to defend the border security.