Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/57810
Title: | 監督式學習演算法之惡意來電偵測 Who's Calling? Malicious Call Detection Using Supervised Learning Algorithms |
Authors: | Ting-Ni Chen 陳亭霓 |
Advisor: | 雷欽隆(Chin-Laung Lei) |
Keyword: | 機器學習,監督式學習,分類,交叉驗證, machine learning,supervised learning,classification,Support Vector Machine (SVM),cross-validation, |
Publication Year : | 2014 |
Degree: | 碩士 |
Abstract: | 隨著科技快速進步,行動電話成了多數人生活中不可或缺的物品。
通訊技術的廣泛流通亦造成電話詐騙等犯罪行為,使用者可能被竊取 個資,甚至造成金錢上的損失。隨著行動電話使用者介面的進步,多 數使用者仰賴電話簿及來電顯示以判斷來電者。然而,若來電號碼未 紀錄於電話簿中,則使用者將無從判斷此來電是否為惡意電話。現今 判斷惡意電話的方法採用黑名單,由使用者回報惡意電話號碼,因此, 判斷的準確率將視其回報者與黑名單的維護而定。 本論文採用監督式學習演算法 (supervised learning algorithm),利用 使用者過去的行為來分類惡意使用者與正常使用者。基於正常的使用 者與惡意使用者有不同行為的假設,試圖讓電腦自動判斷一個號碼的 擁有者是否為惡意使用者,並且分析此分類器 (classifier) 的精確度。 實驗結果顯示,此方式能較傳統方法更早發現惡意電話使用者,讓惡 意來電的自動判斷與偵測成為可能。 With rapid advancement in technologies, mobile phones have gained popularity and become indispensable. The growth of elecommunication has also given rise to malicious calling behaviors where users may encounter theft of identities or even financial losses. Due to the improvement of user interface on mobile phone devices, most mobile phone users rely on caller IDs which link to contact book to identify callers. However, it is often difficult to detect whether an unknown ID is malicious or not without additional information. Recent malicious caller identification establishes blacklists based on user reports. Detecting malicious callers in this fashion proofs to be difficult and inefficient due to the fact that user report is inconsistant and unreliable. Since there might be differences between malicious and benign call patterns, the aim of this study is to automatically predicting whether an unknown ID is malicious or not by observing their past call histories. In this study, we collected phone call histories in two different countries and applied machine learning algorithms to detect whether an unknown ID is benign or malicious. We evaluated the ability of different classifiers and compared the experimental results with conventional blacklist approach. Emperical results suggest that the proposed method is effective and can be a viable approach in detecting malicious calls. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/57810 |
Fulltext Rights: | 有償授權 |
Appears in Collections: | 電機工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
ntu-103-1.pdf Restricted Access | 1.07 MB | Adobe PDF |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.