Here’s the article in markdown format:
Handling Imbalanced Data in Financial Applications using Fuzzy Logic
Imbalanced data is a common problem in financial applications, where one class dominates the others. In this paper, we propose an approach called “weighted scaled dominance” to handle imbalanced data by giving minority classes a good chance when competing with the majority class.
Weighted Scaled Dominance
The weighted scaled dominance approach involves scaling the firing strengths of each rule based on their upper and lower bounds. The following steps are taken:
Steps for Weighted Scaled Dominance
- Scaling: Scale the firing strengths of each rule based on their upper and lower bounds.
- Scaled Confidence: Calculate the scaled confidence, which is the ratio of the number of patterns in the consequent class to the total number of patterns with the same antecedents.
- Scaled Support: Calculate the scaled support, which is the sum of the scaled firing strengths for all rules with the same antecedents and conflicting classes.
- Scaled Dominance: Compute the scaled dominance by multiplying the scaled confidence and scaled support.
- Weighting: Weight the scaled dominance by the average dominance over fuzzy rules with the same antecedent but different consequent classes.
Resolution of Conflicting Rules
To resolve conflicts between rules, we propose replacing multiple conflicting rules with one rule having the same antecedents and a single consequent class that is equivalent to the rule resulting in the highest weighted scaled dominance value.
Prediction Phase
The prediction phase involves calculating the upper and lower membership values for an input pattern based on the produced model. Two possible cases can occur:
Prediction Cases
- Matching Pattern: If the input pattern matches with one of the rules, then the corresponding consequent class is predicted.
- Non-Matching Pattern: If the input pattern does not match with any rule, then a default value or another decision-making mechanism can be used.