1. Train test split the data, rate = 0.25
2. 3-fold cross-validation on the trainning data so that the test size of each split will be equal to the final test size
3. Apply best parameters to test data
4. Compared with default parameters

1. • Dataset: SF 10%
   • Anomaly rate : 4.5%
   • random_state = 1
   • nu : tuned using gridsearch_cv [0.045, 0.18]
   • kernel : default
   • gamma : default
2. • Dataset: SF 10%
   • Anomaly rate : 4.5%
   • random_state : 2
   • nu: tuned using gridsearch_cv [0.045, 0.18]
   • kernel : poly
   • gamma : default
   • degree : default
3. • Dataset: SF 10%
   • Anomaly rate : 4.5%
   • random_state = 3
   • nu : tuned using gridsearch_cv [0.045, 0.18]
   • kernel : default
   • gamma : default

1. • Dataset: SF 100%
   • Anomaly rate : 0.5%
   • random_state = 1
   • contamination : tuned using gridsearch_cv [0.005, 0.2]
   • n_estimator : default
   • max_samples : default
2. • Dataset: SF 100%
   • Anomaly rate : 0.5%
   • random_state = 2
   • contamination : tuned using gridsearch_cv [0.005, 0.02]
   • n_estimator : default
   • max_samples : default
3. • Dataset: SF 100%
   • Anomaly rate : 0.5%
   • random_state : 3
   • contamination : tuned using gridsearch_cv [0.005, 0.02]
   • n_estimator : default
   • max_samples : default

1. • Dataset: SF 10%
   • Anomaly rate : 4.5%
   • random_state = 1
   • contamination : tuned manually [0.045, 0.02]
   • n_estimator : default
   • max_samples : default
2. • Dataset: SF 20%
   • Anomaly rate : 0.5%
   • random_state = 2
   • contamination : tuned manually [0.005, 0.02]
   • n_estimator : default
   • max_samples : default
3. • Dataset: SF 50%
   • Anomaly rate : 0.5%
   • random_state = 2
   • contamination : tuned manually [0.005, 0.02]
   • n_estimator : default
   • max_samples : default
4. • Dataset: SF 70%
   • Anomaly rate : 0.5%
   • random_state : 3
   • contamination : tuned manually [0.005, 0.02]
   • n_estimator : default
   • max_samples : default
5. • Dataset: SF 100%
   • Anomaly rate : 0.5%
   • random_state : 4
   • contamination : tuned manually [0.005, 0.02]
   • n_estimator : default
   • max_samples : default

1. • Dataset: SF 100%
   • Anomaly rate : 0.5%
   • random_state = 1
   • contamination : tuned using gridsearch_cv [0.005, 0.02]
   • n_estimator : default
   • max_samples : default
2. • Dataset: SF 20%
   • Anomaly rate : 4.5%
   • random_state = 2
   • contamination : tuned using gridsearch_cv [0.045, 0.18]
   • n_estimator : default
   • max_samples : default
3. • Dataset: SF 50%
   • Anomaly rate : 10%
   • random_state : 3
   • contamination : tuned using gridsearch_cv [0.01, 0.04]
   • n_estimator : default
   • max_samples : default
4. • Dataset: SF 100%
   • Anomaly rate : 20%
   • random_state : 4
   • contamination : tuned using gridsearch_cv [0.2, 0.8]
   • n_estimator : default
   • max_samples : default

• Tuned algorithms manually starting Week 7 since gridsearch cv proved to not yield a good enough result
• Split the original SA dataset into normal and abnormal dât
• Frac

• Script: notebook/Increasing the number of normal data.ipynb We start with the 1% SA dataset, the data is then split into normal and abnormal data

Week 8 : Worked on

Phan Trung Thành