In machine learning, what we suppose to do if we see a redundant instances and a conflict instances? for example, in the breast cancer data set I noticed
the following redundant instance :(perhaps due to discretizing)
40-49,premeno,25-29,0-2,no,2,right,left_low,no,recurrence-events
40-49,premeno,25-29,0-2,no,2,right,left_low,no,recurrence-events
and the following conflict (diffrent outcomes for the same input data)
30-39,premeno,0-4,0-2,no,2,right,central,no,recurrence-events
30-39,premeno,0-4,0-2,no,2,right,central,no,no-recurrence-events
30-39,premeno,30-34,0-2,no,2,left,left_up,no,no-recurrence-events
30-39,premeno,30-34,0-2,no,2,left,left_up,no,recurrence-events
40-49,premeno,20-24,0-2,no,2,left,left_low,no,no-recurrence-events
40-49,premeno,20-24,0-2,no,2,left,left_low,no,recurrence-events
40-49,premeno,25-29,0-2,no,2,right,left_low,no,recurrence-events
40-49,premeno,25-29,0-2,no,2,right,left_low,no,no-recurrence-events
40-49,premeno,30-34,0-2,no,3,right,right_up,no,no-recurrence-events
40-49,premeno,30-34,0-2,no,3,right,right_up,no,recurrence-events
50-59,premeno,25-29,0-2,no,1,right,left_up,no,recurrence-events
50-59,premeno,25-29,0-2,no,1,right,left_up,no,no-recurrence-events
Should I discard them? to determine the accuracy? or change the outcome to be something like conflict?
For my point of view, it isn't problem to use both Redundant instances and a conflict instances. Certainly, you can (manually) to delete Redundant instances, but it isn't necessary - perhaps, it isn't error, rather (randomly) fully repeated measurements. Same for conflict instances - perhaps, one time this instance has label "+", and another time it has label "-". For example SVM can handle these situations without problems!
@Sasha Yiollin
There is no problem with the redundant instance for both learning and testing but a ware should be considered in evaluation ( this is the point).
For conflict instances, it seems the attributes are not sufficient to make a precise prediction. In fact, I study the combinatorial interaction between the parameter's values and noticed these instances. That is why I ask if it is possible to change the outcome to be Conflict (+ or -) . Or , perhaps, there is a historical dependency to treat the instances for the same person like after or before certain instances.
TQ.
- Badges
- Science topic