Handling Numerical Features on Dataset Using Gauss Density Formula and Data Discretization Toward Naïve Bayes Algorithm
Mochammad Yusa, Ernawati Ernawati, Yudi Agus Setiawan, Desi ADRESWARI · 2020
Naïve Bayes is one of best classifiers in data mining.Naïve Bayes Algorithm either is used in some research areas.Besides having good performances, the algorithm can also handle numerical and categorical data values.This paper presents two ways of treating numerical features as a pre-process before implementing Naïve Bayes algorithm in classifying a dataset.First way is by implementing Gauss Density Formula.In second way, we treat the numerical features to be categorized manually by involving the experts.This study start from collecting data which contains numerical attributes in majority.Then dataset will be treated by using first way and second way.We validate the performance of algorithm by using 10-Fold Cross Validation.The considered performances in this research are accuracy, precision, and recall.The result shows that treating numerical features using Gauss Density Techniques outperforms the treatment by discretizing numerical features of nominal values.First way obtains 80% accuracy, 80,61% of precision average, and 80,41% of recall average value while the second way reaches 65% of accuracy, 63,95% of precision average, and 66,43% of recall average.