1. cost-sensitive classification
1. applicable
Different prediction error costs for different classes
2. How to use
s1. establishment of a cost matrix
below, since the main diagonal represents the prediction is correct, so the cost is 0. elsewhere rely on specific circumstances cost, here we are set to 1.
S2. predicted probability when we multiply a vector select the desired lowest cost forecast
3. When to Use
Depending on the circumstances, suitable and cost matrix used to enhance the effect of time at a suitable
- Ignored during training, consider the forecast period
- Ignore the forecast period, considering the training phase
- Are taken
2. Figure rise
1. How to get increased coefficient?
Figure 2. The rise of meaning?
At the beginning, suppose we have 10,000 samples, the number is 1000. The response
so this presents a writing sample under the theory of linear upward trend, so there will be a straight line.
Since we machine learning methods to give rise coefficient curve thus obtained
3. ROC curve?
1. Why ROC curve?
ROC curve is useful in finding the optimal different classifiers.
2. ROC curve meaning?
Rendering performance classifierRegardless of the type or error distribution costs.
y-axis: number / percentage certainly class. Sensitivity.TP/(TP+FN)
x-axis: Negative Number / percentage of classes. Specificity.FP/(TN+FP)
ROC jagged line depends on the specific content of the test sample
3. Look how ROC curve?
Because we always want to make the program correctly classified, as always tend to choose closer to the y-axis. (Roc therefore say the bigger the better)
For example, A, B are ROC curves two methods.
- Prior A, B cross point, means that a small amount of data, A near the Y-axis, and therefore a method selected from the group A
- After the A, B cross point, means slightly larger amount of data, B near the Y-axis, so selection method B
- When the intersection, in combination
4. Summary
recall=TP/(TP+FN)