Curva ROC figuras esquemáticas Prof. Ivan Balducci FOSJC / Unesp Receiver Operating Characteristic (ROC) Curve DR 1 DR0 c c co c Each threshold c corresponds to a point (FPR, DR) on the X-Y plane. A ROC curve is obtained as c sweeps from - to + . 0 FPR0 FPR 1 2 Lab Tests: What is “Abnormal”? Desempenho True Positive True Negative B False Negative A False Positive TP – Classe é A e classificamos como A TN – Classe é B e classificamos como B FP – Classe é B e classificamos como A FN – Classe é A e classificamos como B The Cut-off Value Trade off • Sensitivity and specificity depend on the cut off value between what we define as normal and abnormal • Assume high test values are abnormal; then, moving the cut-off value to a higher one increases FN results and decreases FP results (i.e. more specific) and vice versa • There is always a trade off in setting the cut-off point Receiver Operating Characteristic (ROC) Curve DR 1 DR0 c c co c Each threshold c corresponds to a point (FPR, DR) on the X-Y plane. A ROC curve is obtained as c sweeps from - to + . 0 FPR0 FPR 1 6 ROC Curve Receiver Operating Characteristic (ROC) Curves Goodness-Of-Fit: Other Measures of Model Performance • ROC (Receiver Operating Characteristic) Curve • Sensitivity and Specificity are dependent on a given cut-point c. • An ROC curve is obtained by plotting sensitivity against (1specificity) for an entire range of possible cut-points. • The area under the ROC curve is a measure of the model’s ability to discriminate between event and non-event in the following fashion: » Among all possible pairs (event, non-event), the proportion of pairs for which the event has higher probability than the corresponding non-event is equal to the area under ROC. ROC Curve • Interpretation: Area Under ROC Curve – If randomly selected pairs of subjects (one with event and one with non-event) are classified in such a way that the subject with higher estimated probability of the event belongs to the event group and the other subject to non-event group, then the proportion of correctly classified such pairs of subjects would be equal to the area under ROC • Generally Accepted Rule: • ROC = 0.5: no discrimination (no better than coin toss) • 0.7 <= ROC < 0.8: acceptable discrimination • 0.8 <= ROC < 0.9: excellent discrimination • ROC > 0.9: outstanding discrimination • Roc area is often used to compare predictive ability of different models; The ROC H1 PTP PTP H0 PFP 1 Decision threshold • The ROC shows the tradeoff between PFP and PTP as the threshold is varied AZ PFP 0 0 1 Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Healthy Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Healthy Sick Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Fals pos= 20% True pos=82% Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Fals pos= 9% True pos=70% Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Healthy Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Healthy Sick Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Fals pos= 20% True pos=82% Developmental characteristics: Cut-points and Receiver Operating Characteristic (ROC) Fals pos= 9% True pos=70% Evaluating the results 1 How can we measure the performance of a feature matcher? 0.7 true # true positives # matching features (positives) positive rate 0 0.1 false positive rate # false positives # unmatched features (negatives) 1 ROC curve (“Receiver Operator Characteristic”) How can we measure the performance of a feature matcher? 1 0.7 true # true positives # matching features (positives) positive rate 0 0.1 false positive rate 1 # false positives # unmatched features (negatives) ROC Curves • • • • Generated by counting # current/incorrect matches, for different threholds Want to maximize area under the curve (AUC) Useful for comparing different feature matching methods For more info: http://en.wikipedia.org/wiki/Receiver_operating_characteristic Curva ROC Cut-off especificidade Área sob curva sensibiidade