Thesis 1 by Shane Delos Reyes

XGBoost

sdelosreyes0421_ — 2024-03-17 13:22:20 UTC

An Effective Cost-Sensitive XGBoost Method for Malicious URLs Detection in Imbalanced Dataset

2024-03-17 13:22:43 UTC

Problem: Imbalanced Class, produced imbalanced ratio between number of positive lables and negative labels is ignored.

Solution: Cost Sensitive XGBoost

Recommendation:

Explore improvement of classifiers in terms of training time and rate of malicious URL identification.
Explore other techniques like explainable AI, reinforcement learning, domain transfer learning, and multi model learning.

XGBoost Advantages

2024-03-17 13:22:48 UTC

compatibility with various programming languages
popularity stems from its exceptional performance, often referred to as the "2P's" - speed of processing (speed is attributed to parallelization) and high-quality outputs
incorporates regularization and auto-pruning to prevent overfitting
handles missing values seamlessly through a sparsity-aware algorithm

Malicious URL Detection

sdelosreyes0421_ — 2024-03-17 13:24:27 UTC

Feature Engineering

sdelosreyes0421_ — 2024-03-17 13:24:38 UTC

Extracting/collecting the features of a URL link using word-based features via word segmentation

Data Imbalance

sdelosreyes0421_ — 2024-03-17 13:24:43 UTC

Solutions based on literatures:

DIA-XGBoost
SMOTE-XGBoost

xgboost with feature engineering

2024-03-17 13:24:49 UTC

xgboost ensembled with other ml models such as random forest or support vector machine

2024-03-17 13:26:16 UTC

Overfitting

2024-03-17 13:28:38 UTC

occurs when a machine learning model learns the details and noise in the training data to the extent that it performs well on the training data but poorly testing data

computational problem

2024-03-17 13:29:14 UTC

be able to handle more complex attacks/ URLs and mas efficient in theory mag classify

Classifying URLS

sdelosreyes0421_ — 2024-03-17 13:31:14 UTC

URL must be classified (ex. spam, malicious, phishing, and benign)

Parallelization

2024-03-17 13:32:17 UTC

Instead of doing one task at a time, multiple tasks are done simultaneously. In the case of XGBoost, when it's working on creating decision trees (a key part of how it learns from data), it doesn't do it all in one go. Instead, it splits the work into smaller parts and works on them at the same time.

URL Redirection Attack Mitigation in Social Communication Platform using Data Imbalance Aware Machine Learning Algorithm

2024-03-17 13:34:32 UTC

Problem: Poor classification accuracy

Solution: DIA-XGBoost

Recommendation:

Present another algo for multiple URL detection and train the model on complicated malicious URLs.
Improve classification and detection.
Improve concept drift problems in detection of malicious URLs.

XGBoost Disadvantages

2024-03-17 13:41:08 UTC

sensitive to:

Noisy Data

2024-03-17 13:49:59 UTC

Outliers

2024-03-17 13:50:07 UTC