Husni’s research while affiliated with Trunojoyo University and other places

Monitoring child development is vital in Indonesia due to its large child population and varying socio-economic and geographical conditions. Malnutrition adversely affects children's growth and development, with ongoing challenges in remote areas despite government efforts. This study addresses the need for accurate nutritional status classification to improve intervention strategies. This study applies the Support Vector Machine (SVM) classification method to analyze and classify nutritional status of toddlers using data from 473 samples collected from health centers in Bangkalan Regency. The classification includes categories such as Good Nutrition, Excess Nutrition, Obesity, and Risk of Excess Nutrition. The SVM model achieved an accuracy of 76% in predicting nutritional status.

Hypertension is a disease caused by increased blood pressure above 140/90 mmHg and is often referred to as "the silent killer" because most sufferers do not realize that they have hypertension, and only realize when complications have occurred. Hypertension is one of the main causes of death worldwide which can be influenced by many factors. In UPT (Integrated Service Unit) PUSKESMAS (Community Health Center) Modopuro, Mojokerto Regency, hypertension is ranked among the top 10 diseases with the most patients. With a fairly high risk of death and an increase in the number of people with hypertension, it is often caused by delays in diagnosis, which must be carried out blood pressure checks by medical personnel at least 2 times with 1 week to establish a diagnosis of hypertension. If hypertension is not treated immediately, it can cause other health conditions such as kidney disease, heart disease, and stroke. Therefore, a system is needed that can be used for the classification of early detection of whether a person has hypertension or not. To overcome these problems, a system was created to classify hypertension using the Backpropagation method. Backpropagation is very effective in helping artificial neural networks learn from mistakes, allowing the system to make more accurate predictions over time. Dataset used in this study is the medical record data of UPT Puskesmas Modopuro patients with 1000 data. The results obtained the best model with a network structure of 7-5-1, learning rate 0.001, and Adam optimizer. With an accuracy of 93.50% and a loss value of 0.0697. While the precision, recall, and f1-score values are 94.00%, 93.00%, and 93.00%, respectively. With good accuracy performance, indicating that the backpropagation model can be applied in hypertension classification.

Backpropagation Neural Network (BPNN) is a suitable method for predicting the future. It has weaknesses, namely poor convergence speed and instability, requiring parameter tuning to overcome speed problems, and having a high bias. This research uses the Random Search hyperparameter technique to optimize BPNN to automatically select the number of hidden layers, learning rate, and momentum. The added accuracy of momentum will speed up the training process, produce predictions with better accuracy, and determine the best architectural model from a series of faster training processes with low bias. This research will predict the local Indonesian cattle population, which is widely developed by people in the eastern part, especially Madura, in 4 types of cattle: sono cattle, karapan cattle, mixed cattle, and breeder cattle. The results of BPNN hyperparameter measurements with the best model show that hyperparameter optimization did not experience overfitting and experienced an increase in accuracy of 2.5% compared to the Neural Network model without hyperparameter optimization. Based on the test results, the BPNN algorithm parameters with a data ratio of 70:30, the best architecture for backpropagation momentum is 6-6-1, with a learning rate of 0.002, momentum 0.3, which has an MSE during testing of 0.1176 on Karapan type Madurese cattle. Tests based on computing time measurements show that the BPNN hyperparameter algorithm stops at 490 iterations compared to regular BPNN. The research results show that the hidden layers, learning rate, and momentum if optimized simultaneously, have a significant influence in preventing overfitting, increasing accuracy, and having better execution times than without optimization.

The COVID-19 pandemic has resulted in significant changes in the education sector. The government issued a policy so that learning must be carried out online from home. This policy became a polemic for teachers and students so that pro and con opinions emerged on social media, especially Twitter. Sentiment analysis of public opinion is an interesting study. Standard classification algorithms such as k-Nearest neighbours, naïve bayes, decision tree, random forest, and support vector machine (SVM) can categorize these opinions in a short time with good accuracy. Many studies show that SVM is more accurate than all other classification methods. SVM works using kernels, including Linear, Polynomial and Radial Basis Functions (RBF) where each kernel requires different parameters. The linear kernel only requires one parameter, namely c (Cost). The RBF kernel requires 2 parameters, c and ɣ (gamma) while the Polynomial kernel uses 2 parameters, c and degrees. SVM does not have default values for these parameters and are based on experience and experimentation. The wider the range of parameters, the more likely the classifier obtains the optimal values. This study tries some parameters values of SVM kernels for text classification based on sentiment. Testing using 5-fold cross validation and confusion matrix show that SVM with a linear kernel provides the best performance with an accuracy of above 84%.

Corn disease has a significant impact on both the food industry and the yield of corn crops since corn serves as a fundamental and essential source of nutrition, especially for vegetarians and vegans. Therefore, ensuring the quality of corn is crucial, and to achieve this, protection against various diseases is necessary. Consequently, there is a pressing demand for an automated method capable of early-stage disease detection and prompt action. However, detecting diseases at an early stage poses a major challenge and is of utmost importance. This research focuses on the development of a classification model for corn stalk images using Random Forest. The model generates fine and coarse features of high quality to capture discriminative, boundary, pattern, and structural information used in the classification process. This research also utilizes the LBP (Local Binary Pattern) method and Color Histogram in the feature extraction process to obtain information related to texture and distinguishing patterns, that are employed in the classification process. Furthermore, the proposed model is evaluated using the corn plant image dataset, which was directly captured by the researcher in Madura, and consists of 3,000 data. The result of this research shows that the utilization of the proposed method can classify and identifying diseases in new data of digital images of corn stalks with an accuracy rate of 99.05%.

Technological developments cannot be denied a very meaningful impact on human life so that what was previously done traditionally is now completely digital, for example conventional news media which has been transformed into an online news portal so that it can still reach its readers. Online news portals provide a lot of up-to-date information, including tourism news, which is an industry that continues to grow and has the opportunity to create new jobs for the community. Currently, to search for tourism news, people only need to type a keyword (query) in the search engine which will then display the latest news about tourism. However, not all tourism news displayed matches what they are looking for, so readers have to re-check which takes a lot of time. Therefore, a tourism news retrieval system that can display the most relevant tourism news to the query is proposed. The Modified Cosine method shows good results in document clustering to bring the inter-cluster distance closer. This study uses the Modified Cosine method and TF-IDF weighting schema to determine the value of precision, recall, and f-measure in calculating the similarity of the query to tourism news documents. The system has been tested using 3 types of queries, with 5 different words each. The test results show that Modified Cosine method obtained best precision value in the test using 5 words in the query and the F-Measure value and the best recall on the 3 words test in query.

The work of finding documents that are relevant to a user's query on an information retrieval system (IRS) is a very interesting study. The relevance of the list of documents returned by the IRS is influenced by the accuracy of the method of calculating the similarity between documents and the determination of the keywords. Many users are difficult to describe their information needs in words. Sometimes the user enters only one or two words that do not reflect the domain of information required. This results in a list of documents were very less relevant to the user's needs. The approach to improve the list of words in the user's query to make it more representative is called Query Expansion. One technique that can be used to expand a query is Pseudo Relevance Feedback. This paper describes the results of research that has been carried out to expand Query using Pseudo Relevance Feedback on an IRS based on the Indonesian version of the Wikipedia dataset, totaling about 450 thousand documents. Calculation of the similarity between the query and the list of tourism news documents uses the cosine similarity, while the weighting scheme for each term uses TF-IDF. The test results show that the pseudo-relevance feedback decreases the precision of the IRS up to 30%. This is due to the failure of the chosen approach to finding the right words to expand the original query. The abstract of articles in Wikipedia is general and is not limited to the tourism domain. The selection of the expansion base dataset is greatly determined by the new query quality and datasets from the same domain are recommended. It is highly recommended that the QE reference dataset is domain specific and filtered before being used as a QE basis.