Stanford University public Class machine learning: Machines Learning System Design | Trading off precision and recall (F score formula: How to balance (trade-off) precision and recall values in a learning algorithm)

In general, the relationship between recall and precision is as follows:1, if the need for a high degree of confidence, the precision will be very high, the corresponding recall rate is very low, 2, if the need to avoid false negative, the recall rate will be very high, the precision will be very low. on the right, the relationship between recall rate and precision ratio is shown in a learning algorithm. It is important to note that no learning algorithm can simultaneously guarantee high precision and recall rate, to high precision or high recall rate, depending on their own needs. In addition, the relationship between the precision ratio and recall rate can be varied, not necessarily the shape of the diagram.

How to choose the precision and recall rate values :

The algorithm presented at the outset has an average of the precision and recall rates, as the following formula Average= (P+R)/2. Obviously, in the three algorithms given, the average of the algorithm 3 is the highest, but it is not a good model by the precision ratio (0.02) and recall (1.0). Therefore, it is not advisable to take an average of this evaluation mode.

It is a useful algorithm to use the F-score algorithm to evaluate both precision and recall rates . The PR of the molecule determines that the precision ratio (P) and recall (R) must be large at the same time to ensure that the F score values are larger. If the precision ratio or recall rate is very low, close to 0, the direct result of the PR value is very low, approaching 0, that is, F score is also very low.

At this point we compare three algorithms, we can find that the algorithm 1 is optimal, and we observed that the algorithm 3 in this formula F score value is the lowest. It is good to note that algorithm 3 is not a good model (precision is too low). Description F score is a good formula for evaluating both precision and recall rates.

Stanford University public Class machine learning: Machines Learning System Design | Trading off precision and recall (F score formula: How to balance (trade-off) precision and recall values in a learning algorithm)