Indriana Hidayah

SARIMA (Seasonal ARIMA) implementation on time series to forecast the number of Malaria incidence

The usefulness of forecasting method in predicting the number of disease incidence is important. It motivates development of a system that can predict the future number of disease occurrences. Fluctuation analysis of forecasting result can be used to support the making of policy from the stake holder. This paper analyses and presents the use of Seasonal Autoregressive Integrated Moving Average (SARIMA) method for developing a forecasting model that able to support and provide prediction number of diasease incidence in human. The dataset for model development was collected from time series data of Malaria occurrences in United States obtained from a study published by Centers for Disease Control and Prevention (CDC). It resulted SARIMA (0,1,1)(1,1,1)12 as the selected model. The model achieved 21,6% for Mean Absolute Percentage Error (MAPE). It indicated the capability of final model to closely represent and made prediction based on the Malaria historical dataset.

Automated microaneurysms (MAs) detection in digital colour fundus images using matched filter

Diabetic retinopathy (DR), one of the most common causes of blindness, is a retinal abnormality caused by high glucose in diabetic patients that leads to micro vascular complications. DR has five levels of severity, i.e. no DR, mild non-proliferative diabetic retinopathy (NPDR), moderate NPDR, severe NPDR and proliferative diabetic retinopathy (PDR). Microaneurysms (MAs), the first sign of NPDR, can be used as a pre-indicator of DR. However, a manual assessment on digital colour fundus images conducted by ophthalmologists is time consuming. This paper introduces a new algorithm for the automated microaneurysms (MAs) detection in digital colour fundus images using matched filter. Generally, the algorithm consists of four phases, namely green band extraction, MAs and blood vessels isolation, MAs and blood vessels detection, and blood vessels removal. To validate the developed algorithm, the results are compared with their ground truths and annotations using ROI based validation. This algorithm obtains an average sensitivity, specificity, accuracy, and false positive number of 91.0603%, 99.9752%, 99.9752% and 256.44 pixels, respectively. This indicates that the proposed algorithm successfully detects microaneurysms and is able to be implemented in a system for DR mass screening purposes.

Semi-supervised learning approach for Indonesian Named Entity Recognition (NER) using co-training algorithm

The problem of utilizing machine learning approach in Indonesian Named Entity Recognition (NER) system is the limited amount of labelled data for training process. However, unlike the limited availability of labelled data, unlabelled data is widely available from many sources. This enables a semi-supervised learning approach to solve this NER system problem. This research aims to design a semi-supervised learning model to solve NER system problem. A semi-supervised co-training learning is used to utilize unlabelled data in NER learning process to produce new labelled data that can be applied to enhance a new NER classi□cation system. This research uses two kinds of data, Indonesian DBPedia data as labelled data and news article text from Indonesian news sites (kompas.com, cnnindonesia.com, tempo.co, merdeka.com and viva.co.id) as unlabelled data. The pre-processing steps applied to analyze unstructured text are sentence segmentation, tokenization, stemming, and PoS Tagging. The results of this pre-process are the NER and its context used as unlabelled data for the semi-supervised co-training process. The SVM algorithm is used as a classi□cation algorithm in this process. 10 Cross Fold Validation is used as the system testing approach. Based on the result of the NER testing system, the precision is 73.6%, the recall is 80.1% and f1 mean is 76.5%.

The Comparation of Distance-Based Similarity Measure to Detection of Plagiarism in Indonesian Text

The accesible loose information through the Internet leads to plagiarism activities use the copy-paste-modify practice is growing rapidly. There have been so many methods, algorithm, and even softwares that developed till this day to avoid and detect the plagiarism which can be used broadly unlimited on a certain subject. Research about detection of plagiarism in Indonesian Language develop day by day, although not significant as English Language. This paper proposes several models of distance-based similarity measure which could be used to assess the similarity in Indonesian text, such as Dice’s similarity coefficient, Cosine similarity, and Jaccard coefficient. It implemented together with Rabin-Karp algorithm that common used to detect plagiarism in Indonesian Language. The analysis technique of plagiarism is fingerprint analysis to create fingerprint document according to n-gram value that has been determined, then the similarity value will be counted according to the same number of fingerprint between texts. Small data text about Information System tested in this case and it divided into four kinds of text document with some modified. First document is original text, second is 50% of original text adding with 50% of another text, third 50% original text modified using sinonym and paraphase, fourth some position of text in original text changed. From the experimental result, cosine similarityshow better performance in generating value accuracy compared to the dice coefficient and Jaccard coefficient. This model is expected to be used as an alternative type of statistical algorithms that implement the n-grams in the process especially to detect plagiarism in Indonesian text.

The performance of GM (1,1) and ARIMA for forecasting of foreign tourists visit to Indonesia

Forecasting can be used for helping the decision-makers to determine the next business strategy to improve the quality of Indonesia tourism such as the improvement of the accommodation facility like transportation and lodging, public services, and promotion to introduce Indonesia tourism objects. This research compared the forecasting performance between GM (1,1) and ARIMA models to determine the best method to forecast the number of foreign tourists visit to Indonesia by using limited data. The data used is the national data of foreign tourists arrival in the airport entrance obtained from the BPS Indonesia in the period of 2002 to 2014. From the result of the forecasting accuracy based on RMSE and MAPE showed that GM (1,1) is smaller than of the ARIMA. It indicates that the performance of GM (1,1) is better than ARIMA to forecast the number of foreign tourists visit. However, it can be concluded that both of the models are able to forecast properly because both of them produce MAPE less than 10%.

Application of J48 and bagging for classification of vertebral column pathologies

Disk hernia and spondylolisthesis are examples of pathologies on vertebral column. These traumas on vertebral column can affect spinal cord capability to send and receive messages from brain to the body systems that control sensor and motor. Therefore, accuracy and timeliness of diagnosis for these pathologies are critical. Hence, a classification system can assist radiologists to improve productivity and the quality of diagnosis. In general, Indonesias public hospitals have many patients, thus, such classification system will be a great benefit. However, research about pathology of skeletal system classification in Indonesia is rare due to the unavailability of numerical database which quantitatively represents the disease. In this research, dataset of vertebral column from UCI Machine Learning was used to develop an optimum classification model. We ensemble decision tree (J48) and bagging as the classification model. Decision tree was chosen as the base learner due to its simplicity and interpretability. In addition, bagging was used to stable the prediction of new test instances. By applying 10-fold cross-validation we calculated true-positive rate (TP rate), false-positive (FP rate), accuracy parameters, and ROC AUC. The results showed that J48 and Bagging has better performance than J48 alone. The quantitative evaluation showed accuracy of J48 and Bagging is 85.1613%, whereas accuracy of J48 was 81.6129%.

Fuzzy Logic Tsukamoto for SARIMAOn Automation of Bandwidth Allocation

The wireless network is used in different fields to enhance information transfer between remote areas. In the education area, it can support knowledge transfer among academic member including lecturers, students, and staffs. In order to achieve this purpose, the wireless network is supposed to be well managed to accommodate all users. Department of Electrical Engineering and Information Technology UGM sets wireless network for its daily campus activity manually and monitor data traffic at a time then share it to the user. Thus, it makes bandwidth sharing becomes less effective. This study, build a dynamic bandwidth allocation management system which automatically determines bandwidth allocation based on the prediction of future bandwidth using by implementing Seasonal Autoregressive Integrated Moving Average (SARIMA) with the addition of outlier detection since the result more accurate. Moreover, the determination of fixed bandwidth allocation was done using Fuzzy Logic with Tsukamoto Inference Method. The results demonstrate that bandwidth allocations can be classified into 3 fuzzy classes from quantitative forecasting results. Furthermore, manual and automatic bandwidth allocation was compared. The result on manual allocation MAPE was 70,76% with average false positive value 56 MB, compared to dynamic allocation using Fuzzy Logic and SARIMA which has MAPE 38,9% and average false positive value around 13,84 MB. In conclusion, the dynamic allocation was more effective in bandwidth allocation than manual allocation.

Work in progress: Application of unsupervised learning method toward student's metacognition assessment

There has been awareness of the importance of metacognitive skill in learning processes, especially for university students, who are required to be more self-regulated. Therefore, monitoring the development of such skill is needed to ensure students achievements. Currently, in classroom learning environments, students metacognition can be observed conventionally by using interview or think-aloud procedures, however, the tasks are tedious and impractical for big number of students. Therefore, an automatic students metacognition assessment/modeling is in progress. This paper proposes a partly automatic metacognition assessment, which is an implementation of unsupervised learning method. This proposed method is proven using a case study in which the dataset was gathered by administering Metacognitive Awareness Inventory (MAI) questionnaire to undergraduate students in our department. Experiment shows that the proposed method could be utilized for automatic assessment of students metacognition. Two groups of students are identified, one that does well in their metacognitive awareness and the other one that needs further guidance and advisory to help them achieve better results and avoid failures.

Design adaptive learning system using metacognitive strategy path for learning in classroom and intelligent tutoring systems

Adaptive learning (AL) is a learning process focusing on students’ personality through of recommendation-based learning and inquiry-based learning. This research intends to create a system capable of giving AL recommendation in the form of learning path (LP) in which LP will then be used as a recommendation for the teacher in constructing a learning process. In this research, LP was obtained through the identification of students’ metacognitive with the use of Metacognitive Awareness Index (MAI) respectively. The system itself was developed using Naive Bayer Classifiers (NBC) and rule-based system to construct the LP. Accuracy of the machine learning for learner style module was of 97.25% and black box test indicated that the system was functionally acceptable. From these results, it can be concluded that the system was functionally acceptable and capable of representing an expert (seeing as it could produce an output that conformed to the expected condition). An expert in education had declared the LP wa...

Performance analysis of Naïve Bayes, PART and SMO for classification of page interest in web usage mining

User interaction with web sites generates a large amount of web access data stored in the web access logs. Those data can be used for e-commerce to conduct an evaluation of possessed website pages as one of the efforts to understand the desires of the user. Through classification techniques in web usage mining, we conducted an experiment to categorize a number of data obtained from the client log files in two groups namely interest page and un-interest page by using the model page interest estimation. The results obtained indicate that SMO algorithm forms a better classifier models with the result accuracy of 95.8904% and this result is higher when compared with two other algorithms. It can be concluded that the SMO algorithm is efficient in performing classification for this case.