Author : Mustakim, Assad Hidayat, Zuliar Efendi, Aszani, Rice Novita, Eplia Triwira Lestari
Publish : Journal of Theoretical and Applied Information Technology. 2018. Vol.96. No 13
Indonesia’s maritime area is twice the size of its archipelago, with an area of 5.9 million km2. Based on the United Nations Convention on the Law of Sea (UNCLOS 1982). Indonesia is also the second largest fish producing country in the world with fish catch of 6 million tons in 2014 based on the latest data from the Food and Agriculture Organization (FAO). The fish catching process requires the role of vessels suited to the existing water conditions, one of which has robust resilience to the state of the Indonesia sea. Thus, it is necessary to study the classification of aquatic types on Indonesian fishing vessels to determine the impact that will occur on the vessel. This research performs classification process using Naïve Bayes Classifier and Probabilistic Neural Network (PNN) algorithm. Accuracy result got in Naïve Bayes Classifier algorithm using RapidMiner tool is equal to 48%. While for PNN algorithm, experiment with three different spread values yield an accuracy of 68% for spread value 0.1, 78% accuracy for spread value 0.01 and the last experiment is the value of spread of 0.001 produce 100% accuracy. Therefore, in this study it is known the classification using PNN algorithm is better than Naïve Bayes Classifier.
The results show the accuracy of Naïve Bayes Classifier algorithm using RapidMiner tool is equal to 48%. While for PNN algorithm the three different spread values of 0.1 yield accuracy by 68%, 78% for spread value 0.01 and last experiment that spread 0.001 yield 100% accuracy were experimented. Therefore, in this study it can be concluded that the classification using PNN algorithm is better than Naïve Bayes Classifier and k-Nearest Neighbor. Data sharing technique using K-Fold Cross Validation with 70% and 30% model can only be applied by using small data, if applied to large data, either record or attribute, will experience higher complexity and drastic decrease in accuracy. In this research some experiments have been conducted using multiples of 100 data or 43.600 record with 5 multiplication attributes or 25 attributes used in this research which only yield accuracy of 47%. Thus increasingly the amount of data and attributes will decrease its accuracy as well. Therefore, it can be concluded that the less data and attributes used then the accuracy will be higher.
The disadvantage of this research is that the amount of data is only 436 data and only has 5 attributes, for further research other neural network methods are proposed such as LVQ, Backpropagation and Perceptron to get better results.