UNP Journal of Statistics and Data Science

Classification of Harvest - Non Harvest in Rice Plant Image Using Convolutional Neural Network Algorithm

2024-08-09T07:02:50+00:00

The Area Sample Framework (ASF) survey is an area based survey carried out by direct observation of sample parts whose locations have been determined. Every month ASF officers take photos of observation results using an Android based cellphone, where the results of the photos will be classified manually by supervision officers and sent to a central server for processing. The large amount of rice plant image data included can hinder officers in classifying rice growth phases. Therefore, to speed up the classification process, the Convolution Neural Network (CNN) method is used. In this research, the CNN model built consists of 3 convolution layers, 3 pooling, ReLU and Sigmoid activation functions, with several other parameters such as batch size and epoch value. The training results show that the accuracy value for the training data is 92.86% with an epoch value of 120. Meanwhile, the accuracy value for the validation data is 69.01%. Model evaluation shows a precision value of 21.34% and a recall value of 32.20%. This shows that the CNN model has poor performance in predicting harvest and non-harvest in rice plant images.

Markov Chain Model Application for Rainfall Pattern in Padang City

2024-08-09T07:00:53+00:00

Rainfall is a natural phenomenon that includes climate variables and is observed every time in every place. Daily rainfall data is a time series data, which is random. It is a data transfer from one time to another which can be expressed as a state of light, medium, heavy or very heavy rainfall intensity. Rainfall prediction is needed for people's lives and supports the economy. In addition, rainfall prediction is an anticipation of prevention if high rain intensity will occur in a long time. One of the rainfall prediction methods that can be used is the stochastic process approach. Markov chain is part of the stochastic process that can be used for prediction of rainfall at the present time based on one previous time. The focus of this research is the application of Markov Chains for rainfall prediction. Through Markov chains, long-term opportunities for rainfall phenomena are obtained. This study will look at the rainfall pattern of Padang City using Markov chains and also to predict rainfall in Padang City. The results of predicting the weather conditions of Padang City with any rainfall conditions today are 36.9% for the chance of no rain tomorrow, 46% for the chance of light rain tomorrow, 10% for the chance of moderate rain tomorrow, 5.3% for the chance of heavy rain tomorrow, and 1.8% for the chance of very heavy rain tomorrow.The results of this study are expected to be a recommendation for parties directly involved in taking preventive measures due to rainfall.

Grouping Potential Forest and Land Fires Areas in Sumatera Island Based on Hotspot Using CLARA Method

2024-08-12T11:28:17+00:00

Sumatera Island is one of the areas with the potential for forest and land fires in Indonesia. Sumatra Island has the largest oil palm plantation in Indonesia. The vast land area of oil palm plantations in Indonesia can increase the risk of fires due to land expansion by burning. In addition, the burning of peatlands in Sumatra can exacerbate the impact of forest and land fires. Forest and land fires on the island of Sumatra that occur every year can cause various negative impacts, indicating the need for countermeasures and prevention efforts to minimize the impact of forest and land fires. Hotspots can be used to detect fires in a region and help with prevention and countermeasures to reduce the impact of land and forest fires. Clustering the hotspot data allows one to obtain information on the presence of a fire in a given area as well as its potential status high, medium, or low. The clustering method used is the CLARA method. The CLARA method is a clustering method that breaks the dataset into groups. The advantages of the CLARA method are robust to outliers and effective for large data sets. The results of this research show that the CLARA method can be used for hotspot clustering with a silhouette coefficient of 0.53 in the use of 2 clusters. The analysis of the clustering results shows that cluster 1 is a cluster with low fire potential while cluster 2 is a cluster with high fire potential.

Grouping Potential Forest and Land Fires in Indonesia Based on Hotspot Distribution Using CLARANS Method

2024-08-12T06:41:06+00:00

Kebakaran hutan/lahan merupakan bencana yang sering terjadi di beberapa negara di dunia. Peristiwa ini mendapat perhatian lebih dari pemerintah karena menimbulkan banyak kerugian seperti ekonomi, ekologi dan sosial. Indonesia merupakan negara dengan tingkat bencana kebakaran hutan/lahan yang tinggi, hal ini menjadikan Indonesia sebagai negara penyumbang pencemaran terbesar ketiga di dunia. Sehingga diperlukan upaya penanggulangan sejak dini, salah satu upaya yang dapat dilakukan adalah dengan memanfaatkan data titik api dengan melakukan klasifikasi wilayah yang berpotensi terjadinya kebakaran hutan/lahan. Kebakaran hutan/lahan ditandai dengan terdeteksinya data titik api oleh satelit yang terindikasi sebagai titik api. Pada penelitian ini parameter yang digunakan adalah lintang, bujur, kecerahan, keyakinan dan FRP (fire power radiative) dengan menerapkan metode CLARANS. CLARANS merupakan varian dari algoritma k-medoid dan juga merupakan pengembangan dari algoritma sebelumnya, seperti PAM dan CLARA untuk menangani jumlah data yang lebih besar dan tahan terhadap outlier. Hasil penelitian ini menunjukkan bahwa penggunaan metode CLARANS dapat digunakan untuk proses clustering data hotspot dengan hasil koefisien siluet sebesar 0,896 pada penggunaan 2 cluster dengan jumlah data sebanyak 12,287. Hasil cluster menunjukkan bahwa cluster 1 termasuk dalam potensi tinggi dengan kecerahan rata-rata 340K dengan kepercayaan rata-rata 95% dan cluster 2 termasuk dalam potensi sedang dengan kecerahan rata-rata 327 K.

Classification of Dropout Rates in West Sumatra Using the Random Forest Algorithm with Synthetic Minority Oversampling Technique

2024-08-13T07:05:56+00:00

This study aims to classify school dropout rates in West Sumatra Province using the Random Forest algorithm with the Synthetic Minority Oversampling Technique (SMOTE). Based on 2021 data from the Ministry of Education, Culture, Research, and Technology (Kemdikbudristek), the dropout rate in West Sumatra is above the national average. Despite efforts to reduce dropout rates, results remain suboptimal. Therefore, this study seeks to identify the causes of student dropouts and compare the performance of the Random Forest algorithm with and without SMOTE. The study uses the 2021 dropout data from West Sumatra, which has a significant class imbalance. SMOTE is applied to balance the data. The dataset is split into training and testing sets in an 80%:20% ratio, and parameter tuning is performed to optimize mtry and the number of trees (ntree). The model is evaluated using a confusion matrix to compare performance. The results show that Random Forest with SMOTE outperforms the version without SMOTE, with improvements in precision, recall, and F1-score. The presence of the biological mother ( ) is identified as the most significant factor influencing student dropouts, based on the Mean Decrease Gini value. The study concludes that using SMOTE in the Random Forest algorithm helps reduce classification bias and enhances the model's ability to detect students at risk of dropping out.

Analysis of Sumatera Island's Population Based on Age Groups Using Profile Analysis

2024-08-12T01:48:43+00:00

The distribution of the population in each province according to age groups in Sumatra Island has tended to change over time. Therefore, an analysis is needed to provide a comparative overview of the characteristics between the populations of each province with different age groups. This analysis can help to understand the variations in these characteristics in relation to the population. Profile analysis is a technique within multivariate analysis of variance that can be used to examine the differences between two or more populations, where each population is influenced by several treatments (variables) tested. This method has been applied in various fields, including government, to understand the characteristics of specific regions. This study aims to identify the characteristics of the population in each province on the island of Sumatra based on sixteen age groups. Sumatra is one of the largest islands in Indonesia, comprising ten provinces. In this research, profile analysis is utilized to compare the population profiles of each province in Sumatra based on the sixteen age groups. Based on the profile parallelism test, it was found that the profiles of the ten provinces are not parallel, indicating differences in the average population numbers or trend patterns among the provincial profiles in Sumatra based on age groups. Further testing using Tukey's HSD method was conducted to compare each pair of provinces based on specific age groups. The testing revealed that there are significant differences in several provinces in Sumatra for each age group.

Vector Error Correction Model to Analyze the Impact of Exchange Rates and Money Supply on Inflation in Indonesia

2024-08-12T01:50:14+00:00

This study analyzes inflation in Indonesia in relation to the influence of exchange rates and the money supply (M2), which pose challenges in controlling inflation amidst rapid economic growth. Data from the Ministry of Trade of the Republic of Indonesia (Kemendag) were used to investigate the relationship between exchange rates and the money supply (M2) on inflation using the Vector Error Correction Model (VECM). The results indicate that in the short term, inflation tends to decrease towards stability, with a strong exchange rate capable of reducing inflation, while an increase in the money supply slightly raises inflation. However, in the long term, inflation demonstrates a strong self-correction mechanism, with the influence of exchange rates and the money supply becoming limited. This model proves effective in forecasting inflation for the period from March to August 2024, with a Mean Absolute Percentage Error (MAPE) of 19.59%.

Markov Chain Application to Daily Rainfall Data in Semarang City

2024-08-09T14:02:29+00:00

Rainfall is a measure of the amount of water that falls on the earth's surface in a given period of time. High rainfall can cause flooding in certain areas, while low rainfall can leave areas vulnerable to drought. Semarang City is one of the largest cities in Java Island that is often hit by floods. Efforts can be made to anticipate the risk of flooding, one of which is by studying the pattern of rainfall. This study will determine the chances of rainfall transition in Semarang City in steady state conditions using Markov chains. The results are expected to be used to anticipate the risk of flooding in Semarang City. The probability of daily rainfall transition in Semarang City in each state for the next period of time is 90.5% chance of staying in the light rain state, 7.97% chance of staying in the medium rain state and 1.50% chance of staying in the heavy rain state.

The Mapping of Economic Growth Indicators in West Sumatra Province Using Multiple Correspondence Analysis

2024-08-08T05:44:16+00:00

Economic growth is a key factor in sustainable regional development. This study employs Multiple Correspondence Analysis (MCA) to explore the relationships among economic growth indicators in the districts/cities of West Sumatra Province. Data from 2022 provided by the Central Statistics Agency are used to analyze economic growth indicators, including Gross Regional Domestic Product (GRDP) at Constant Prices (X₁), Human Development Index (X₂), Labor Force Participation (X₃), Domestic Investment (X₄), Government Expenditure (X₅), and Balance Fund Allocation (X₆). The results of MCA reveal complex relationships among these variables, with the first and second dimensions explaining approximately 44.43% of the data variance. The MCA plots visualize clusters of districts/cities based on their economic characteristics. From these plots, it is concluded that there are disparities in economic growth indicators in West Sumatra Province, with 11 districts/cities requiring special attention to achieve equitable and sustainable economic growth. This study contributes to a deeper understanding of regional economic disparities in West Sumatra Province and their relevance to more targeted and sustainable development policies.

Application of Extreme Learning Machine Algorithm (ELM) in Forecasting Inflation Rate in Indonesia

2024-08-08T05:44:01+00:00

One indicator to determine the economic stability of a country can be seen from the inflation rate of a country. Inflation is an economic symptom in the form of a general increase in prices or a tendency to increase the prices of goods and services in general and continuously. In an effort to anticipate the impact of inflation in the future, an analysis is needed to find out how the development of the inflation rate is by forecasting. Extreme Learning Machine (ELM) is a feed-forward artificial neural network (ANN) algorithm with one hidden layer called Single Hidden Layer Neural Networks (SLFNs). Based on the research, forecasting the inflation rate in Indonesia using the Extreme Learning Machine algorithm obtained the best architecture (12,48,1) with a MAPE value of 11%. These results show good forecasting because the resulting MAPE is relatively low.

K-Medoids Cluster Analysis for Grouping Provinces in Indonesia Based on Agricultural Households ST2023

2024-08-12T06:31:33+00:00

Agriculture plays a crucial role in Indonesia's national development, providing essential resources such as raw materials, household income, and contributing significantly to Gross Domestik Product (GDP). According to the 2023 Agricultural Census (ST2023), there has been an increase in the number of Agricultural Household Enterprises (RTUP) across various agricultural subsectors. However, the welfare of agricultural entrepreneurs remains low, with 48.68% of poor household heads working in this sector. Therefore, an analysis is needed to understand the patterns and characteristics of RTUPs in each province. This study aims to cluster the provinces in Indonesia based on the number of Agricultural Household Enterprises (RTUP) using K-Medoids cluster analysis. K-Medoids, an extension of K-Means, was chosen for its ability to handle outliers by using medoids as cluster centers instead of means. The research utilized data from the 2023 Agricultural Census, covering 38 provinces and eight variables representing different agricultural subsectors. The optimal number of clusters was determined using the Elbow method, resulting in four distinct clusters. The findings revealed that Cluster 1 consists of 12 provinces with moderate RTUP numbers, Cluster 2 includes 23 provinces with low RTUP numbers, Cluster 3 comprises one province with high RTUP numbers, and Cluster 4 contains two provinces with very high RTUP numbers. The cluster validation using the Davies-Bouldin Index (DBI) yielded a value of 0.722, indicating that the clustering results are optimal.

DBSCAN Method in Clustering Provinces in Indonesia Based on Crime Cases in 2022

2024-08-01T04:30:16+00:00

Based on Central Statistics Agency 2023 data, in 2022 there was a significant increase in the number of crime cases in Indonesia compared to 2021, from 239,481 cases to 372,965 cases. The increase in the number of criminal acts occurred along with community activities that began to loosen up after the Covid-19 pandemic. The types of crimes that occur in Indonesia themselves vary, ranging from murder, theft, drug-related crimes, and others. This research will cluster provinces in Indonesia based on crime cases with certain types of crimes in 2022 using the Density Based Spatial Clustering of Applications with Noise (DBSCAN) method. The results of the study are expected to help the government and police in an effort to deal with crime in Indonesia. Clustering using the DBSCAN method produces 2 clusters with a silhouette coefficient value of 0,68. The resulting cluster is cluster 0 with noise category consisting of 5 provinces with a high number of crime cases, while cluster 1 consists of 29 provinces with a low number of crime cases.

Application of Multivariate Adaptive Regression Splines for Modeling Stunting Toddler on The Island of Java

2024-08-13T06:53:09+00:00

Stunting is a chronic nutritional problem experienced by toddlers, characterized by a shorter body height compared to children their age. The aim of this research is to model and determine the factors that influence Stunting on The Island of Java using Multivariate Adaptive Regression Spline (MARS). MARS is a modeling method that can handle high-dimensional data. The results of this study show that the best MARS model is a combination (BF=24, MI=3, and MO=2) with a minimum GCV value of 0.9475. Based on the model, the factors that significantly influence Stunting on the island of Java are babies receiving complete basic immunization (X4), babies getting exclusive breastfeeding (X3), pregnant women getting K4 (X1), and pregnant women getting TTD (X2). The level of importance of each variable is 100%, 81.64%, 60.38%, and 43.90%. Based on research results, babies receiving complete basic immunization is the variable that most influences stunting on The Island of Java in 2021.

Mixed Geographically Weighted Regression Modeling of Gender Development Index in Indonesia

2024-08-09T13:39:31+00:00

The Gender Development Index (GDI) is one of the primary measures of gender equality in the field of human development. Indonesia's GDI statistics for 2023 show the development gap between men and women. Using Mixed Geographically Weighted Regression (MGWR), a blend of regression and Geographically Weighted Regression (GWR) models, to identify the factors influencing GDI is one approach to closing the gap. The results showed that when it came to value selection using the Akaike Information Criterion (AIC), the MGWR model outperformed the GWR model. Population with health complaints and adjusted per capita expenditure were found to be globally influential factors, while female participation in parliament, open unemployment rate, and labor force participation rate were found to be locally influential factors by the MGWR model with Adaptive Kernel Bisquare weights.

Implementation of the Fuzzy C-Means Clustering Method in Grouping Provinces in Indonesia based on the Types of Goods Sold in E-commerce Businesses in 2022

2024-08-12T02:34:36+00:00

The internet facilitates e-commerce by enabling efficient transactions and building consumer trust. With internet users in Indonesia reaching 204 million in 2022, it is crucial to Cluster provinces based on the types of goods and services sold online to design effective marketing strategies. The Fuzzy C-Means (FCM) method is used for Cluster analysis, allowing objects to have different membership degrees in multiple Clusters and providing accurate Cluster center placement. This study applies Fuzzy C-Means to Cluster 34 provinces in Indonesia based on the sale of goods/services in e-commerce in 2022, aiming to provide insights into market preferences and assist companies in developing more effective strategies. The results show that the method forms two Clusters. By evaluating standard deviation values and ratios, Fuzzy C-Means proves effective in Clustering provinces in Indonesia based on e-commerce sales data. Cluster validation reveals a standard deviation ratio of 0.14, indicating clear and significant Cluster separation.

Comparison of Extreme Learning Machine and Holt Winter’s Exponential Smoothing Methods in Railway Passengers Forecasting

2024-08-09T13:45:40+00:00

Forecasting the number of passengers on the Pariaman Express train is an activity that is considered to have the potential to help PT KAI in maximizing passenger service facilities and comfort. It is estimated that the number of train passengers in Indonesia will always increase along with the increasing population of Indonesia. The high interest of users of this mode of transportation can be seen from historical data that continues to increase every year. PT KAI (Persero) as a single train transportation provider company needs to have several strategies in providing and meeting passenger needs every day. In the study of forecasting the number of passengers on the Pariaman Express train using the Holt Winters exponential smoothing method and one of the artificial neural network methods, namely the extreme learning machine. The purpose of this study was to determine the comparison of the accuracy values of the forecast results produced by the two methods, and to find out which method is good to use in this forecast. The data used is data on the number of Pariaman Express train passengers from 2021-2023. The results of the study show that the comparison of the accuracy values of the forecasting of the number of train passengers shows that the Holt Winter's and ELM methods have error values above 10%, meaning that the Holt Winter's and ELM methods are good at forecasting for 4 periods. Holt Winter's has a MAPE value of 17.10% and ELM has a MAPE value of 20%.

Factors Evaluation of Indonesia Human Index Development in 2023 Using PLS-SEM Method

2024-08-09T13:59:48+00:00

The human development index (HDI) is a measure of the success of development in a country. Indonesia as a developing country in 2022 has an HDI value that ranks 112 out of a total of 193 countries in the world. This indicates that there is an urgent need for evaluation in increasing the HDI value in Indonesia which leads to an increase in the quality of human development. The evaluation can be done using the Structural Equation Modeling-Partial Least Square (SEM-PLS) analysis method. With 34 Indonesian provinces as observations, there are three dimensions as variables analyzed in this paper, namely economy, education, and health. These variables are analyzed based on each indicator variable. The results of the analysis show that in the economic variable, the influential indicators are the Open Unemployment Rate, GRDP per Capita at Constant Prices, and Average Wage per Hour Worker. Then in the education variable, the influential indicators are the School Participation Rate Age 7-12, the School Participation Rate Age 13-15, the Pure Enrollment Rate for Elementary/Middle School/Package A, the Pure Enrollment Rate for Junior High School/MTs/Package B, and the Pure Enrollment Rate for Senior High School/SMK/MA/Package C. Furthermore, in the health variable, there are indicators of the Percentage of Households by Province and Source of Adequate Drinking Water, and the Percentage of Ever-Married Women Aged 15-49 Years whose Last Childbirth Processed in a Health Facility which affect the value of HDI in Indonesia in 2023.

Comparison of Linear Discriminant Analysis with Robust Linear Discriminant Analysis

2024-08-12T02:36:44+00:00

Discriminant analysis is a multivariate method for dividing things into discrete groups and assigning new objects to existing categories. A discriminant function, which is a linear combination of independent variables used to categorize things into two or more groups or categories, is the result of discriminant analysis. The independent variables in a linear discriminant analysis must be multivariate normally distributed, and the covariance matrices for each group must be equal. In linear discriminant analysis, it is also essential to identify outliers because their existence in the data set can undermine the assumptions made by the method and lead to incorrect classification results. Therefore, in discriminant analysis, handling outliers with robust approaches is required. One such robust method in discriminant analysis is the Minimum Covariance Determinant (MCD), which is highly effective in dealing with outliers and relatively easier to apply compared to other robust methods. The aim of this study is to compare the classification results of linear discriminant analysis with robust linear discriminant analysis on the dataset of diabetes patients at RSUD Padangsidimpuan in 2023. The results obtained from this dataset indicate that linear discriminant analysis achieved an accuracy of 85,71%, while robust linear discriminant analysis achieved an accuracy of 80,95%. These findings suggest that the use of liniar discriminant analysis and robustt linear discriminant analysis can yield different results depending on the characteristics of the data and the number of outliers in the dataset.