From: An overview of recent distributed algorithms for learning fuzzy models in Big Data classification
Algorithm | Strengths | Weaknesses |
---|---|---|
Chi-FRBCS-BigData | The first distributed algorithm proposed in the literature for learning a fuzzy model in big data classification | Employs a local search, thus the structure of the final model depends on how data chunks are generated |
Adopts a single reducer for fusing the rules generated by a distributed mapping stage | ||
Generates a large number of rules | ||
Generally achieves accuracies lower than the comparison algorithms | ||
CHI_BD | Global search: unlike Chi-FRBCS-BigData, employs a global search, thus the structure of the final model does not depend on how data chunks are generated | Generates a large number of rules |
Generally achieves accuracies lower than the comparison algorithms | ||
DFAC-FFP | Includes a fuzzy discretization algorithm | Generates a large number of rules |
The generated models are very accurate | The input variables may be partitioned with a large number of fuzzy sets, thus the interpretability of the fuzzy partitions may be low | |
DPAES-RCS | Optimizes concurrently the rule bases and the parameters of the fuzzy sets | Adopts a pre-fixed number of fuzzy set for each input variable |
Generates solutions characterized by good trade-off between accuracy and interpretability | Is very slow with respect to the other algorithms (it is based on evolutionary optimization) | |
Even the most accurate solutions are characterized by a reduced number of rules | Â | |
DPAES-FDT-GL | Adds to the strengths of the PAES-RCS algorithm the capability of optimizing also the number of fuzzy sets for each attribute | Is very slow with respect to the other algorithms (it is based on evolutionary optimization) |
Multi-way FDT | Includes a fuzzy discretization algorithm | Is characterised by a low interpretability of the final models because of the large number of rules generated |
Is very fast for generating the models | Â | |
The fuzzy classification models are very accurate | Â | |
\(FMDT_{l}\) | Adds to the strengths of the Multi-way FDT algorithm the capability of reducing the model complexity | The final models are still characterised by a low interpretability because of the large number of rules |