Clustering refers to reducing selected features involved in determining the clusters. Raw data might come with a lot of features, including unimportant ones. A hybrid similarity measure (discovered in 2014) used in selecting features can be improvised as it might select all the attributes, including insignificant ones. This paper suggests Fuzzy Lambda-Max to be used as a feature selection method since Lambda-Max is normally used in ranking of alternatives. A set of AIDS data is used to measure the performance. Results show that Fuzzy Lambda-Max has the ability to determine criteria weights and ranking the criteria. Hence, feature selection can be done by choosing only the important criteria.
The aim of the study was to determine the success factors of oral cancer susceptibility prediction using fuzzy models. Three fuzzy prediction models including fuzzy logic, fuzzy neural network and fuzzy linear regression models were constructed and applied to a Malaysian oral cancer data set for cancer susceptibility prediction. The three models’ prediction performances were evaluated and compared. All the three fuzzy models were found to have 64% prediction accuracies for 1-input and 2-input predictor sets. However, when the number of input predictor set was increased to 3-input and 4-input, both fuzzy neural networks’ and fuzzy linear regression’s prediction accuracies increased to 80%, while fuzzy logic prediction accuracy remains at 64%. Fuzzy linear regression model was found to have the capability of quantifying the relationships between input predictors and the predicted outcomes and also suitable for small sample size. Fuzzy neural network model on the other hand, handles ambiguous relationship between variables well but lacks the ability to describe input-output association. The third model, fuzzy logic, is easy to construct but highly dependent on human expert-input. The outcome of this study is a computer-based prediction tool which can be used in cancer screening programs.