I am using the Gaussian Mixture Model model and Gaussian Model to classify anomaly data.
When I am using the raw dataset the f1score and accuracy are relatively less and when I translate ( multiply each data by some constant(say 32000) I am getting high f1score and accuracy?
why this is happening? Is the above model is sensitive to data variance?
Firstly, I suggest that you could normalize or standardize the data (https://scikit-learn.org/stable/modules/preprocessing.html#standardization-or-mean-removal-and-variance-scaling). Then, you could train the model. In addition, I suggest you could classify anomaly data by calculating their Cook’s distance (Fig. 3 in Chem. Mater. 2018, 30, 4031−4038). This method is usually simple, effective and time-consuming
If your data is big enough, it should not be an issue, due to the law of large numbers. On the other hand, if your data is small, you should consider some data preprocessing work to identify the influential observations responsible for the variance explosion.
- Badges
- Science topic