Jurnal Online Informatika
Not a member yet
276 research outputs found
Sort by
Comparison of Template Matching Algorithm and Feature Extraction Algorithm in Sundanese Script Transliteration Application using Optical Character Recognition
Get PDFThe phenomenon that occurs in the area of West Java Province is that the people do not preserve their culture, especially regional literature, namely Sundanese script, in this digital era there is research on Sundanese script combined with applications using Feature Extraction algorithm, but there is no comparison with other algorithms and cannot recognize Sundanese numbers. Therefore, to develop the research a Sundanese script application was made with the implementation of OCR (Optical Character Recognition) using the Template Matching algorithm and the Feature Extraction algorithm that was modified with the pre-processing stages including using luminosity and thresholding algorithms, from the two algorithms compared to the accuracy and time values the process of recognizing digital writing and handwriting, the results of testing digital writing algorithm Matching algorithm has a value of 87% word recognition accuracy with 236 ms processing time and 97.6% character recognition accuracy with 227 ms processing time, Feature Extraction has 98% word recognition accuracy with 73.6 ms processing time and 100% character recognition accuracy with 66 ms processing time, for handwriting recognition in feature extraction character recognition has 83% accuracy and 75% word recognition , while template matching in character recognition has an accuracy of 70% and word recognition has an accuracy of 66%
Detection of Fraudulent Financial Statement based on Ratio Analysis in Indonesia Banking using Support Vector Machine
Get PDFThis study proposes the use of ratio analysis-based features combined with the SVM classifier to identify fraudulent financial statements. The detection method used in this study applies a data mining classification approach. This method is expected to replace the expert in forensic accounting in identifying fraudulent financial statements that are usually done manually. The experimental results show that the proposed classifier model and ratio analysis-based features provide more than 90% accuracy results where the optimal number of features based on ratio analysis is 5 features, namely Capital Adequacy Ratio (CAR), (ANPB) to total earning assets and non-earning assets (ANP), Impairment provision on earning assets (CKPN) to earning assets, Return on Asset (ROA), and Return on Equity (ROE). The contribution of the study is to complement the research of fraudulent financial statements detection where the classifier method used here is different compare to other research. The selection of banking cases in Indonesia is also unique in this research which distinguishes it from other research because the financial reporting standards in each country can be different.Â
Diagnosis of Types of Diseases in Cassava Plant by Bayes Method
Get PDFThis research was conducted to implement the Bayes algorithm in an expert system to diagnose types of diseases in cassava plants. The research data was taken from the Binjai City Agriculture and Fisheries Office in 2018. The expert system was built based on the web, where the application was built using the PHP programming language and MySQL DBMS. The results showed that the Bayes algorithm can be used in expert system applications to diagnose types of cassava plant diseases. In the Bayes algorithm, the knowledge base is taken from the data of the amount of data from cassava plants that suffer from disease, so the results of diagnosing cassava plants are based on existing data. Therefore, the more patient data that is used as a knowledge base, the better the diagnosis results are given
Analysis and Implementation Machine Learning for YouTube Data Classification by Comparing the Performance of Classification Algorithms
Get PDFEvery day, people around the world upload 1.2 million videos to YouTube or more than 100 hours per minute, and this number is increasing. The condition of this continuous data will be useless if not utilized again. To dig up information on large-scale data, a technique called data mining can be a solution. One of the techniques in data mining is classification. For most YouTube users, when searching for video titles do not match the desired video category. Therefore, this research was conducted to classify YouTube data based on its search text. This article focuses on comparing three algorithms for the classification of YouTube data into the Kesenian and Sains category. Data collection in this study uses scraping techniques taken from the YouTube website in the form of links, titles, descriptions, and searches. The method used in this research is an experimental method by conducting data collection, data processing, proposed models, testing, and evaluating models. The models applied are Random Forest, SVM, Naive Bayes. The results showed that the accuracy rate of the random forest model was better by 0.004%, with the label encoder not being applied to the target class, and the label encoder had no effect on the accuracy of the classification models. The most appropriate model for YouTube data classification from data taken in this study is Naïve Bayes, with an accuracy rate of 88% and an average precision of 90%
Geographic Information Systems for Crime Prone Areas Clustering
Get PDFCrime is one of the problems that is quite complicated and very disturbing to the community. Crimes can occur at different times and places, making it difficult to track which areas are prone to such actions. K-means algorithm is used to cluster prone areas and Geographic Information System is used to map crime-prone areas. Web-based application is developed with the PHP programming language. The data used is quantitative data in the form of the number of crimes committed and the coordinates of the cases. The attributes of the crime used consist of five parameters: theft, mistreatment, rape, women and child protection cases and fraud. The results of this study are clustering areas into 3 cluster and mapping prone areas that is safe area, safe enough area and prone area. From the overall crime data for 2019 in Purwakarta district, it was found that 68.75% was safe, 18.75% was quite safe and 12.5% was prone area
Genetic Algorithm to Optimize k-Nearest Neighbor Parameter for Benchmarked Medical Datasets Classification
Get PDFComputer assisted medical diagnosis is a major machine learning problem being researched recently. General classifiers learn from the data itself through training process, due to the inexperience of an expert in determining parameters. This research proposes a methodology based on machine learning paradigm. Integrates the search heuristic that is inspired by natural evolution called genetic algorithm with the simplest and the most used learning algorithm, k-nearest Neighbor. The genetic algorithm were used for feature selection and parameter optimization while k-nearest Neighbor were used as a classifier. The proposed method is experimented on five benchmarked medical datasets from University California Irvine Machine Learning Repository and compared with original k-NN and other feature selection algorithm i.e., forward selection, backward elimination and greedy feature selection. Experiment results show that the proposed method is able to achieve good performance with significant improvement with p value of t-Test is 0.0011
Implementation of Image Enhancement Algorithm for Image Forensics using Mathlab
Get PDFThe purpose of this journal is to explain the implementation of the image enhancement algorithm for image forensics. Image Forensic deals with the types of digital evidence in the form of digital image files. One of the most commonly used digital devices in providing digital evidence for forensic analysis is CCTV (Closed-Circuit Television). CCTV images have a low quality such as noise, blur, lack of light intensity, etc., so that the image must be enhanced so that forensic analysis can be done. To enhance image quality, an application is needed by applying the image enhancement algorithm. The algorithm applied to the application is a Low Pass Filter to increase low pixel intensity, High Pass Filter to increase high pixel intensity, Median Filter to replaces the original pixel value with the pixel center value of the image, Mean Filter to replaces the original pixel value with a value the average pixel of the image, the Gaussian Filter for reducing noise in the image, the Wiener Filter to reduce blur in the image, the Histogram Equalization spreads the image histogram value, Contrast Stretching to stretch the contrast intensity in the image and Bicubic Interpolation to increase the image size and resize the image. In this study, the application was built using MATLAB and the testing process for each algorithm was based on Timing-Run, MSE and PSNR parameters. From the test, the average MSE value is 1058.512083 and the PSNR value is 541.61875 dB, which means that the resulting image has a fairly high level of similarity and the average time needed to process the algorithm for the image is 0.114627915 seconds
Exploiting Web Scraping for Education News Analysis Using Depth-First Search Algorithm
Get PDFOnline news is one source of data that is always up to date and provides information or factual data. The search engine is one of the features for users to be able to enter keywords based on the expected category quickly. The development of education in Indonesia makes it essential to discuss, in this study using unstructured data in online news with the keyword Education included as a parameter, and adding search methods in the field of Artificial Intelligence so that the data becomes more accurate. Data that used here was from online news, namely CNN Indonesia, Detikcom, and Liputan6. Using Python Programming with depth-first search method (DFS), when compared with the results data for relevant news. Web erosion using DFS will be very helpful in searching because this method can check the date data was sent and then track the destination URL. Of the three online media sites, Detikcom produces the highest monthly data yielding an average of 885 news about education. At the same time, Liputan6 has the least amount of data on average, 28 news per month, but the data obtained are very relevant compared to Detikcom and CNN Indonesia
Customer Loyality Segmentation on Point of Sale System Using Recency-Frequency-Monetary (RFM) and K-Means
Get PDFIt is no doubt that the development of the business world has been progressive. Point of sale is one of the many system used as a means of payment in various existing businesses, especially in heterogeneous markets. The activity of transactions between Point of Sale Systems and Customers occur in the business world. Keep in mind also that one of the main factors of business success, is from customers. There is the need of an attractive strategy and certainly it will be to increase the income and assets of a business. To know that, this research will explore the behavior of customer which is based marketing, through RFM Method (Recency, Frequency and Monetary). The case of this study is in Goldfinger Store. It will do segmentation and also use data mining technique to do clustering by using K-Means with result of loyal and potential customer. The results of segmentation using RFM (Recency, Frequency, Monetary) and K-Means methods have produced multiple clusters by dividing them into groups
Analyzing and Forecasting Admission data using Time Series Model
Get PDFProblems that will be faced by higher education institutions, especially in the phase of new student admissions. Careful planning and strategies are needed in dealing with the process of admission of new students. The data for planning can be obtained using the forecasting method. The time series forecasting model is used to get forecasting data. Forecasting data is used for the decision making process. The data of new student admissions obtained is 3-period data (2017 - 2019). The data obtained is stationary. Because the data is stationary, the data does not need differentiation. The data obtained also has a sufficient correlation value, and has a loop on the 7th lag. Before making an application, a test is performed to find a time series model that is suitable for admission data. The tested models are the ARIMA model and the Autoregression model. In testing the forecast timespan, the ARIMA model gets a smaller error value in almost all tests. In the Cross-validation method, the ARIMA Model also gets a smaller RMSECV or MAECV value than the AR model. The ARIMA model was chosen to be implemented into the application. The auto_arima algorithm is used so that applications can adapt to different data. The ARIMA model is implemented into a prediction application using the Python programming language. Application development uses Django as a web-based web application framework. Bootstrap is used to create application interfaces. the result from forecasted data is acceptable for short period