This project focuses on anomaly detection using machine learning models applied to the UGR'16 dataset, specifically the July 2016 dataset. For our analysis, we used a sample of the dataset, selecting one out of every 50 lines for efficiency.
This project aims to detect anomalies within the provided dataset using three different machine learning models, namely Isolation Forest, One-Class SVM, and XGBoost. The results indicate the performance of each model on both the training and test sets. These metrics serve as a benchmark for understanding the effectiveness of each method in identifying anomalies in the data.
| Model | Train Set Balanced Accuracy | Train Set Precision | Train Set Recall | Train Set R2 | Train Set F1 Score | Test Set Balanced Accuracy | Test Set Precision | Test Set Recall | Test Set R2 | Test Set F1 Score |
|---|---|---|---|---|---|---|---|---|---|---|
| Isolation Forest | 49.36% | 0.42% | 0.74% | -1.79 | 0.53% | 49.43% | 0.48% | 0.86% | -1.82 | 0.61% |
| One-Class SVM | 71.23% | 2.28% | 91.94% | -38.98 | 4.45% | 68.78% | 1.93% | 85.98% | -43.43 | 3.77% |
| XGBoost | 99.99% | 97.73% | 100.00% | 0.98 | 98.85% | 99.73% | 97.02% | 99.50% | 0.96 | 98.25% |
The dataset used in this project is the July 2016 dataset from the UGR'16 dataset, sampled for efficiency in the analysis.