Jurnal Pilar Nusa Mandiri

Lexicon-Based and Naive Bayes Sentiment Analysis for Recommending the Best Marketplace Selection as a Marketing Strategy for MSMEs

Hoiriyah Hoiriyah, Helva Mardiana, Miftahul Walid, Aang Kisnu Darmawan — Fri, 01 Sep 2023 00:00:00 +0700

MSMEs (micro, small, and medium enterprises) play an essential role in the Indonesian economy, contributing to 60% of the country's GDP (gross domestic product), creating jobs, and increasing non-oil and gas exports. However, MSMEs in Indonesia face various challenges, including access to technology, digital marketing tools, financial resources, limited market distribution, and low technological literacy. Marketplaces provide an essential marketing channel for MSMEs to increase their competitiveness and sales. Sentiment analysis can assist businesses in making informed decisions about which marketplace to use to increase customer satisfaction. Apart from the importance of the marketplace for MSMEs in Indonesia, research on sentiment analysis for marketplace recommendations is still minimal. Therefore, this study aims to analyze six popular marketplaces in Indonesia using Lexicon-based and naïve Bayes research methods to provide the best marketplace recommendations for MSME marketing. The results showed that Blibli.com had the highest accuracy, followed by Tokopedia, Tiktokshop, Lazada, Shopee, and Bukalapak. Blibli.com received positive reviews with 96.33%, followed by Tokopedia with 95.25%, Tiktokshop with 94.61%, and Lazada with the highest accuracy. 94.22%, Shopee 92.18%, and Bukalapak 89.57%. This research has two significant contributions. First, making a scientific contribution by applying a combination model of lexicon-based and naïve Bayes to analyze market sentiment in Indonesia Second, offering a practical contribution by providing recommendations to MSME actors and policymakers in choosing the best marketplace for MSMEs marketing purposes in Indonesia. By utilizing the recommended marketplace, MSMEs can optimize their marketing strategy and increase their competitiveness in the digital marketplace.

The Prediction Of Product Sales Level Using K-Nearest Neighbor and Naive Bayes Algorithms (Case Study : PT Kotamas Bali)

Aris Setiawan Miha Djami, Nengah Widya Utami, A. A. Istri Ita Paramitha — Fri, 01 Sep 2023 00:00:00 +0700

PT Kotamas Bali is a company that operates in tableware and kitchenware, where every sale is sold at various counters. Product sales are permanently printed and entered in sales reports, and there are problems such as the fact that the product is sold very much and it is difficult to see the rate of sale of most products, lots, and not lots. Then, to see the level of product sales, it is necessary to use data mining techniques with the method of Knowledge Discovery in Databases to predict the purchase rate of products using the two algorithms K-Nearest Neighbor and Naïve Bayes. The purpose of this research is so that PT Kotamas Bali can see the sales rate of each product sold so that there is no accumulation of goods and more focus on the most marketed products. These two algorithms result in different accuracy on the 90:10 data split, where the K-Nearest Neighbor algorithm successfully predicted the sales rate of the product with a 99% accuracy rate and was categorized as an excellent classification. The Naïve Bayes algorithm failed to make predictions with an accuracy of only 54% and was classified as a failure classification. ROC performance results on the K-Nearest Neighbor algorithm with an AUC value of 99% and the Naïve Bayes algorithm with an AUC of 74%. K-Nearest Neighbor managed to obtain the highest accuracy, while the Naïve Bayes algorithm failed to conduct classification.

Disease Detection of Rice and Chili Based on Image Classification Using Convolutional Neural Network Android-Based

Rudi Muslim, Zaeniah Zaeniah, Ardiyallah Akbar, Bahtiar Imran, Zaenudin Zaenudin — Fri, 01 Sep 2023 00:00:00 +0700

The current development of machine learning makes it easier for humans to obtain information, especially from images. The presence of processing assistance from machines can increase the accuracy of the information provided to further convince the recipient of the information. Rice and chili farmers in Indonesia have experienced many disease attacks from several types of plant diseases. Not many farmers understand and are good at guessing the diseases that attack their rice and chili plants. So many rice and chili farmers experienced crop failure. This research aims to build a disease-detection system for rice and chili plants based on Android-based image classification. The machine learning method used is Convolutional Neural Network (CNN) with the Mobile Net version one model combined with the Sequential CNN and Tensor Flow Lite models. The results of the transfer learning evaluation on the Mobile Net version 1 model and the sequential CNN model obtained training accuracy of 0.88% with a loss of 0.34%, validation accuracy of 0.84% with a loss of 0.40%, and testing accuracy of 86% with a loss of 43%. Each uses batch 69 of the total training data stopping at epoch 30 from epoch 100. The results of field testing on the application of rice and chili disease detection on 20 images of rice and chili plants can detect Rice Neck Blast disease with a probability of 75% to 100% and Rice Hispa with a probability of 97% to 100%. It can also detect chili plant diseases such as Chili Yellowish with a probability of 83%, Chili Leaf Spot with a probability of 99%, Chili Whitefly with a probability of 91% to 95, Chili Healthy with a probability of 78% to 99%, and Chili Leaf Curl with a probability 75 to 76%. The probability obtained varies according to how likely damage is to rice and chili plants. CNN with the Mobile Net version one model and the Sequential model can extract and classify images so that it has maximum information processing capabilities. This research can make it easier to help farmers identify diseases that attack their rice and chili plants.