The geographical constrain of small islands associated with fisheries commodities, is due to the lack of insufficient infrastructure and socio-economy characteristics, in addition to the nature of fisheries commodities where the durability and inconsistency of quality and quantity of the particular region become very vulnerable to uncertainty of supply chain. This study aims to formulate a strategy to mitigate the existing effect of uncertainty supply chain in Southeast Maluku Regency. The data mining approach with naïve bayes was used to determine the clusters of areas with high to low interdependency in villages and sub-districts. In addition, the House of Risk (HOR) was used to identify the cause of risks and provided the most effective and efficient alternative strategy to mitigate the uncertainty of the fisheries' supply chain sector. Data were collected from 28 villages to obtain 322 datasets by using the purposive sampling technique and expert assessment method for the data mining process, and HOR, respectively. The result show that the village location, resources, and fishermen's ability to sell and buy their gear were the main determinant factors of interdependency. Furthermore, there were, at least, three clusters of interdependencies with the integration of a web-based system to help spread information to fishermen and aquaculture farmers. In conclusion, policymakers need to be more adaptable to the use of information technology to efficiently improve the welfare and performance of coastal communities, aside from constantly enhancing the region's infrastructure.

The Covid-19 pandemic which has spread across all countries, including Saudi Arabia, has caused the government to create limited curfew policies in the country that affected the economy. This policy has given rise to a new trend in society, namely the habit of shopping online. The trend of purchasing online via e-commerce increases. However, people's opinions and attitudes towards this trend vary. Therefore, this research was conducted with the aim of determining the subjectivity of public opinion or sentiment on the e-commerce activities using probability and statistical approaches, i.e.: the Naïve Bayes (NB) and Support Vector Machine (SVM) classifiers. Three experimental scenarios of dataset splitting for training and testing; 90%:10%; 80%:20%; and 70%:30%. The comparison of accuracy values was carried out using an automatic labeling method. Experimental results show that the 70%:30% split scenario provides the best result, with 89% of accuracy, 99.7% of Precision, 88% of Recall and 93.5% of F1-score for the SVM classifier.