UNP Journal of Statistics and Data Science

Application of Partial Least Squares and Robust Approaches in Discriminant Analysis for High-Dimensional Data

2025-07-25T02:06:01+00:00

Classical discriminant analysis, namely linear discriminant analysis and quadratic discriminant analysis, is generally known to suffer from singularity problems when exprerienced with high-dimensional data and is not robust to outliers that make the data not multivariate normally distributed. This research focuses on investigating the classification performance of discriminant analysis on high-dimensional data by applying two approaches, namely the Partial Least Square (PLS) dimension reduction approach as a solution to high-dimensional data and a robust approach with the Minimum Covariance Determinant (MCD) estimator technique that is robust to outliers. The data used for this study is Lee Silverman Voice Treatment (LSVT) data. PLS forms five optimal latent variables that represent predictor variable information. Based on the assumption test of covariance homogeneity between groups, the test statistic value is greater than the chi-square table or the p-value is smaller than the significance level, which means that the assumption is unfulfilled, so quadratic discriminant analysis is applied. The evaluation results showed that the quadratic discriminant analysis analysis model with the MCD approach on the PLS transformed data was able to achieve 81% accuracy, 71% precision, 86% recall, and 77% F1-score. These values indicate that both approaches are able to maintain the efficiency of discriminant analysis classification performance on high-dimensional and multivariate non-normally distributed data.

PayPal Usage in Indonesia with K-Nearest Neighbor Algorithm

2025-07-25T02:24:13+00:00

The development of information and digital technology has had a significant impact on the financial sector. In Indonesia, digital payment technologies such as PayPal, Gopay, Shopeepay, OVO, and DANA have become an integral part of the modern payment system. Since the implementation of the national electronic clearing system, RTGS, and ATMs in 2005, transactions have become increasinglyconvenient. This study analyzes user sentiment toward PayPal in Indonesia to understand user experience and provide insights for service development, marketing strategies, and brand reputation management. Review data from the PayPal app was collected from Google Plat via web scrapping and processed to yield 597 clean data points. Initial sentiment was categorized into positive, neutral, and negative, wordcloud visualization displayed positive and negative sentiment, while neutral sentiment was analyzed numerically. Automatic labeling was performed using the NLTK library based on rating values, above 3 positive, below 3 negative, and exactly 3 neutral. The results showed 146 positive reviews, 451 negative reviews, and a few neutral reviews. Sentiment classification using the K-Nearest Neighbor (K-NN) method yielded adequate accuracy, indicating that PayPal's acceptance in Indonesia is largely influenced by users' negative experiences. These findings provide a foundation for developing strategies to improve service quality and update PayPal's operational policies in the Indonesian market.

Panel Data Model Selection and Significant Determinants of New Family Planning Participants in West Sumatra

2025-07-25T02:10:03+00:00

Population issues in Indonesia are not limited to poverty, urbanization, population explosion, or high birth rates, but also include how small families can improve and maintain their quality of life. The main objective of the Family Planning program is to create happy and prosperous families with an ideal number of children. The West Sumatra Provincial Health Office report (2023) emphasizes that increasing the number of new family planning acceptors is an important priority to support the success of maternal, child, and family planning health programs, in line with the 2020–2024 RPJMN policy direction. Therefore, this study aims to develop the best panel data model and identify the factors that significantly influence the number of new family planning participants in West Sumatra Province. The secondary data used were obtained from the Statistics Indonesia (BPS) publication entitled West Sumatra Province in Figures from 2021 to 2024. The observation units in this study were 19 districts/cities in West Sumatra Province with a time series from 2020 to 2023. The results indicate that the best-selected model is the random effect model, with the number of couples of reproductive age proven to have a significant effect on the number of new family planning participants. The R-square value of 53.11% indicates that the model can explain 53.11% of the variation in the dependent variable, while the remaining 46.89% is influenced by other factors not included in the model.

Comparison of Expectation-Maximization (EM) Algorithm and Kmeans for District/City Clustering in West Sumatera Province Based on Breadfruit Production

2025-07-25T02:09:18+00:00

Breadfruit (Artocarpus altilis) is an important food source that is highly nutritious and plays a strategic role in West Sumatra Province. However, challenges such as pests, diseases and marketing constraints affect its cultivation and productivity. This study employed K-means and expectation-maximisation (EM) clustering methods to categorise regions according to their breadfruit cultivation characteristics. The elbow method identified three optimal clusters for K-means and seven for EM. Evaluating the quality of the clusters using the silhouette coefficient produced values of 0.47 and 0.37 for EM and K-Means respectively, indicating that EM produced tighter, more distinct clusters. These results suggest that EM is a more effective method for describing the variation in breadfruit production in West Sumatra. With this in mind, the research is expected to inform strategic decision-making aimed at increasing the productivity and added value of breadfruit crops in the area..

Nonparametric Regression with Local Polynomial Kernel on Relationship Between Schooling Years and Unemployment Rate in Banten

2025-06-18T03:09:18+00:00

The Open Unemployment Rate (TPT) is a key indicator in assessing the economic performance of Banten Province. One of the factors suspected to influence TPT is education, which is measured by the average years of schooling. This study aims to analyze the relationship between the average years of schooling and TPT using the Local Polynomial Kernel Nonparametric Regression method for the period 2017–2024. This method was chosen for its flexibility in modeling nonlinear relationships without requiring strict assumptions about the data. The optimal bandwidth parameter for smoothing was determined using the Direct Plug-In (DPI) method through the dpill function in the R software. The results show that the nonparametric model has a coefficient of determination (R²) of 0.2841, which is higher than that of the Ordinary Least Squares (OLS) linear regression model, which only reached 0.1710. This indicates that the nonparametric approach is better at capturing the complex relationship between education and unemployment. However, the low R² values in both models indicate the presence of other factors that influence the unemployment rate, such as economic conditions, labor market structure, and education policy. Therefore, increasing the average years of schooling alone may not be sufficient to significantly reduce the unemployment rate. More comprehensive policies are needed, such as job skill enhancement, vocational training, and economic strategies focused on job creation. The findings of this study are expected to provide useful insights for policymakers in formulating more effective strategies to address unemployment in Banten Province.

Forecasting Inflation Rate in Indonesia Using Autoregressive Integrated Moving Average Method

2025-07-25T02:14:43+00:00

Inflasi merupakan salah satu indikator penting untuk menilai stabilitas ekonomi suatu negara. Peningkatan inflasi yang terus menerus akan memperlambat pertumbuhan ekonomi. Oleh karena itu, prakiraan tingkat inflasi yang akurat penting untuk perencanaan ekonomi jangka menengah hingga panjang. Penelitian ini dilakukan untuk meramalkan tingkat inflasi di Indonesia selama 12 periode mendatang, yaitu dari Januari 2025 hingga Desember 2025. Penelitian ini menggunakan metode ARIMA, karena model ARIMA bersifat fleksibel terhadap semua jenis pola data deret waktu, meskipun data tersebut bersifat non-stasioner. Hasil penelitian menunjukkan bahwa ARIMA (2,0,2) merupakan model terbaik dengan nilai akurasi MAPE sebesar 25,21%. Model ini dapat memprediksi tingkat inflasi yang stabil di Indonesia selama 12 periode mendatang, dengan rata-rata sebesar 1,861%. Hasil ini menunjukkan bahwa kenaikan harga umum barang dan jasa di Indonesia selama periode tersebut akan stabil tanpa fluktuasi, yang merupakan tanda positif bagi stabilitas makroekonomi dan daya beli masyarakat.

Comparison of Nadaraya-Watson Method with Local Polynomial in Modeling Human Development Index and Poverty Relationship in Java

2025-06-18T03:12:36+00:00

Poverty remains a critical issue in Indonesia, with the number of poor people reaching 24.06 million in September 2024. The Human Development Index (HDI), which indicates the level of human resource quality, is one of the factors influence poverty. This analysis focuses on the correlation involving HDI also this number of poor people in districts/cities in Java Island by comparing two kernel regresokesion methods, namely Nadaraya-Watson Estimator and Local Polynomial Estimator. Nonparametric regression was chosen thus it does not necessitate this presumption of a certain form of connection among variables, so it is more flexible in capturing complex relationship patterns. Secondary data from Statistics Indonesia (BPS) in 2024 was used in this study. Initial exploration shows, the data distribution does not have a clear pattern, so nonparametric methods are more suitable for use. Modeling is done using the optimal bandwidth obtained through the dpill function in R software. The analysis results show that the local polynomial estimator produces smoother regression curves and lower MSE values. In addition, comparison of different polynomial degrees shows that higher polynomial degrees tended to improve model performance. Among the tested polynomial degrees, the local polynomial with degree five (p=5) produced the lowest MSE value and the highest coefficient of determination. Therefore, the local polynomial estimator with degree 5 is the best method for modeling the relationship between the HDI and poverty levels in Java in 2024

Aplication Algorithm Learning Vector Quantization for Classification of Hypertention in Padang Laweh Health Center

2025-08-05T01:20:43+00:00

Hypertension is a health condition characterized by blood vessel disorders, in which there is a chronic increase in blood plessure of 140/90 mmHg. There are several factors that influence hypertension, including unhealthy eating patterns, lack of physical activity, smoking, stress and excess weight. Hypertension does not show clear symptoms, but it has the potential to cause other diseases such as heart failure, stroke, and premature death. Therefore, a study was conducted to classify the risk of hypertension based on hypertension diagnoses at the Padang Laweh Health Center, Dharmasraya Regency, using the Learning Vector Quantiazation (LVQ) Algorithm. The advantage of LVQ is its ability to achieve high accuracy in processing data with numerous numerical and categorical features. The analysis results show that the use of the Learning Vector Quantization Algorithm on the test data produces very good accuracy, namely 95.17% correct classification of hypertensive patients

Process Capability Analysis of OPC Cement Production Using Statistical Process Control and IMR Method: Blaine Test Evaluation

2025-07-25T02:15:02+00:00

The main challenge in cement production at PT Semen Padang is maintaining consistent product quality, particularly the fineness of cement particles measured by the Blaine test. Variations in raw materials and the production process can cause fluctuations in quality, which affect the performance of the final product. Therefore, it is crucial to monitor and control process stability and capability to consistently meet product specifications. Based on the Statistical Process Control (SPC) analysis using Individuals and Moving Range (I-MR) control charts on 28 observations of Ordinary Portland Cement (OPC) Blaine values from February 2025, one out-of-control point was detected on the Moving Range chart between observations 16 and 17, indicating a significant variation. However, all points on the Individuals chart remained within control limits, suggesting that the individual process values were still under control. After revising the outlier data, the process was confirmed stable. Process capability analysis showed a Cp value of 2.17 and a Cpk value of 1.98, indicating that the production process is not only statistically stable but also highly capable of meeting quality specifications. Therefore, despite some variation between data points, the cement production process at PT Semen Padang can be considered stable and capable. Nevertheless, periodic evaluations are recommended to maintain consistent product quality and provide strategic recommendations for the Quality Assurance division in implementing data-driven quality control.

Comparison of Kernel and Spline Nonparametric Regression (Case Study: Food Security Index of Jambi Province 2023)

2025-08-06T02:33:27+00:00

Food security is one of the issues that plays an important role in national development, especially in regions with varying levels of economic welfare such as Jambi Province. One of the main factors affecting food security is food expenditure, which reflects the economic capacity of households to access food. The complex and non-linear relationship between Food Security Index (FSI) and Food Expenditure requires a flexible modeling approach in the analysis.

This study aims to compare the performance of nonparametric regression Kernel ans Spline regression methods, namely the Nadaraya-Watson Estimator (NWE) and Local Polynomial Estimator (LPE) for Kernel Regression as well as Smoothing Spline and B-Spline for Spline Regression. The analysis was conducted using secondary data obtained from the Food Security and Vulnerability Map (FSVA) of 2023, with a total of 141 subdistricts in Jambi Province. The response variable is the Food Security Index (FSI), while the predictor variable is Food Expenditure. Model evaluation was conducted using the Mean Squared Error (MSE) and the coefficient of determination (R²).

The results showed that the NWE method had the best performance with the smallest MSE value of 24.47690 and the highest R² value of 0.3332, meaning that approximately 33.32% of the variation in FSI could be explained by Food Expenditure. The LPE method showed nearly comparable performance, while Smoothing Spline and B-Spline exhibited higher prediction error rates. Therefore, the NWE method can be recommended as an effective nonparametric regression approach for modeling the relationship between food expenditure and food security.

Comparison of Nadaraya-Watson and Local Polynomial Methods in Analyzing the Relationship Between Consumer Price Index and Inflation in South Kalimantan

2025-07-25T02:07:40+00:00

This study compares the performance of two nonparametric regression methods, namely Nadaraya-Watson and Local Polynomial, in analyzing the relationship between the Consumer Price Index (CPI) and inflation in South Kalimantan Province. Nonparametric approaches were chosen for their greater flexibility in capturing nonlinear relationships that conventional parametric models may fail to explain. The data were obtained from the Central Statistics Agency (BPS) for the period from January 2022 to December 2024, with missing values in the inflation variable handled through mean imputation. The optimal bandwidth was selected using the direct plug-in method (dpill).Visually, the Nadaraya-Watson method produced a more fluctuating curve that is highly sensitive to local variations, while the Local Polynomial method yielded a smoother and more stable curve. Quantitatively, the Local Polynomial method demonstrated better performance with lower MSE (0.1839), MAE (0.3507), and a higher R² (0.1811) compared to Nadaraya-Watson. These findings indicate that the Local Polynomial method is more effective in balancing curve flexibility and stability. This study also addresses a methodological gap by highlighting the relevance of nonparametric approaches in regional economic analysis. Future research is encouraged to explore alternative bandwidth selection methods and different kernel functions to improve estimation accuracy.

Inflation Prediction in Indonesia Using Extreme Learning Machine and K-Fold Cross Validation

2025-08-05T01:21:45+00:00

Inflation rate forecasting is an important aspect in supporting economic policies and price control by the government. This study aims to evaluate the performance of the Extreme Learning Machine (ELM) algorithm in forecasting the inflation rate in Indonesia and provide inflation prediction results for 2025. The data used is historical data on Indonesia's inflation rate for the period 2003–2024. The analysis process begins with data normalization to ensure a uniform scale, followed by data partitioning using 10-Fold Cross Validation. The ELM model was built with 30 hidden neurons, a sigmoid activation function, and a regularization parameter of 0.8. The test results show that the ELM algorithm has superior performance. This is evidenced by the average MAPE value of 1.71%, RMSE of 0.0359, and coefficient of determination (R²) of 0.9833, indicating very high accuracy. The inflation prediction for January to December 2025 is in the range of 1.517%–1.761%, with an average approaching 1.663%, indicating a relatively stable pattern throughout the year. Based on these results, the ELM algorithm can be used as an effective alternative method for forecasting time series data, particularly in the context of inflation. This research is expected to serve as a reference for the government in establishing inflation control policies and for other researchers interested in applying artificial intelligence models to economic analysis.

Forecast Accuracy Comparison Between Holt’s Method and the Box-Jenkins Approach: The Case of Madiun City Labor Force Participation Rate

2025-08-05T01:17:09+00:00

Pacitan district recorded the highest Labor Force Participation Rate (LFPR) in Eastern Java Province. Meanwhile, Madiun city which is one of the largest cities in East Java Province, is ranked only 34 out of 39 cities in 2023. This condition raises concern for the local government, perticularly the Department of Manpower,in ensuring that the productive-age population can be optimally absorbed into the labor market. The LFPR is categorized as time series data, thus forecasting method are required to estimate its future trends.This Study compares the performance of the Double Exponential Smoothing Holt (DESH) method and the Autoregressive Integrated Moving Average (ARIMA) Box-Jenkins approach in forecasting the LFPR of Madiun City. The empirical result show that the ARIMA (1,0,1) model provides better accuracy compared to DESH. The forecasting result indiacte that the LFPR of Madiun City is project to reach 67,19% in 2024, 67,20% in 2025, and 67,21% in 2026, with Mean Squared Error (MSE) of 14,48; Root Mean Square Error (RMSE) of 3,80 and Mean Absolute Percentage Error (MAPE) of 4,75%. These finding are expected to serve as reference for future research and practical input for policymakers in formulating strategies to improve labor LFPR in Madiun City.

Applications of Panel Data Analysis on Human Development Index Indicators in Districts/Cities of Lampung 2022 – 2024

2025-08-05T01:21:07+00:00

This paper aims to identify the determinants affecting the Human Development Index (HDI) in Lampung Province, Indonesia, during the periode 2022-2024 using panel data regression. Lampung consistenly ranks among the provinces with the lowest HDI scores in Sumatera, indicating developmental disparties across regions. The research employs secondary data from 15 districts/cities and includes variables such as life expectancy, expected years of schoolingm mean years of schooling, and expenditure per capita. Panel data regression models fixed effect, random effect, and common effect were evaluated using chow, hausman, and lagrang multiplier tests to select the most approriate model. The random effect model was chosen, supported by a high R-Squared value of 92,71% indicating strong explanatory power. The analysis found that life expectancy and mean years of schooling significantly influence HDI, while expected years of schooling and expenditure per capita were not statistically significant in this model. The analysis shows that ensuring equal opportunities in health and education significantly contributies to better human development. Future research is recomended to incorporate qualitative approaches and more recent variables to enrich the analysis.

Grouping Regencies/Cities in West Sumatra Province Based on People’s Welfare Indicator Using Biplot Analysis

2025-08-05T01:19:31+00:00

The level of community welfare is a crucial reflection of the success of development in a region. Welfare is assessed based on eight aspects: poverty, employment, education, housing, consumption patterns, health, population, and other social factors. In West Sumatra Province, the level of community welfare still requires improvement across all indicators. The determination of community welfare levels can be achieved by reviewing all dimensions based on the linear relationships between districts/cities, thereby providing insights into the indicators that still need enhancement. This effort can assist the West Sumatra Provincial Government in formulating regional policies and programs for equitable distribution and improvement of community welfare across all districts/cities. The data used in this study are secondary data obtained from the West Sumatra Provincial BPS website in 2024. The grouping of districts/cities was conducted using Principal Component Analysis based on singular value decomposition biplot analysis. The analysis results formed four groups with distinct characteristics of community welfare indicators. The groups that need to be prioritized for improvement are groups 1 and 3, which exhibit low levels of community welfare. Group 2 consists of districts/cities with high community welfare characteristics in terms of population, education, and housing. Meanwhile, group 4 includes districts/cities with high community welfare characteristics regarding consumption patterns, poverty, and labor indicators.

Factors Affecting Households Program Keluarga Harapan Recipients in West Sumatra: Binary Logistic Regression Analysis

2025-08-11T02:13:43+00:00

Poverty is still a complex issues in Indonesia. Poverty rate in West Sumatra province has increased over the past 3 years. One of the government's initiatives to address poverty is the Program Keluarga Harapan (PKH), which is a social protection program that provides conditional cash transfers to poor and vulnerable Keluarga Penerima Manfaat (KPM) on condition that they are registered in the Data Terpadu Kesejahteraan Sosial (DTKS). Although PKH has a positive impact on poverty alleviation and enhanced access to health, education, and social welfare, the implementation still faces major challenges such as data inaccuracies, particularly in targeting accuracy. Therefore, an analysis is needed to determine the factors that significantly affects PKH recipient households in West Sumatra Province. This research used variables from the DTKS variable group contained in SUSENAS 2024 using two stages one phase stratified sampling method with 11,600 observations consisting of 1,790 receiving PKH and 9,810 not receiving PKH. The dependent variable is PKH recipient status (Yes = 1, no = 0). Data were analyzed using binary logistic regression with a significance level of 5%. Based on the results of the analysis, it can be concluded that floor area of the house, age of the household head, household size, education level of the household head, and floor material of the house have a significantly effect on PKH recipient households. Household size has the most influence on PKH receipt with a 40,3% probability of receiving PKH.

Applying Robust Spatial Autoregressive Model to Analyze the Determinants of Open Unemployment in West Java

2025-07-25T02:08:26+00:00

Open unemployment is a critical macroeconomic challenge in developing regions like West Java, Indonesia, where spatial disparities and data anomalies complicate traditional analysis. This study addresses these limitations by employing a Robust Spatial Autoregressive (RSAR) model with M-Estimator, integrating spatial dependence and outlier resilience to enhance estimation accuracy. Using 2024 district-level data from Indonesia’s Central Bureau of Statistics (BPS) and Open Data Jabar, the research examines determinants such as labor force participation, education, and regional GDP. The methodology begins with Ordinary Least Squares (OLS) to identify initial predictors, followed by spatial diagnostics (Moran’s I, Lagrange Multiplier tests) to confirm spatial autocorrelation. A customized Queen contiguity weight matrix captures neighborhood effects, while robust M-Estimation mitigates outlier distortions. Results reveal that the RSAR model achieves superior explanatory power (R² = 0.8626) compared to OLS and standard Spatial Autoregressive (SAR) models, with labor force participation (X₄) emerging as a significant negative predictor of unemployment. Spatial effects (ρ = 0.337) though modest, underscore the importance of inter-regional dynamics. The study concludes that RSAR offers a more reliable framework for regional labor analysis, combining spatial rigor with robustness against data irregularities. Policy-wise, the findings advocate targeted interventions to boost labor participation and address localized disparities, emphasizing the need for spatially informed, outlier-resistant methodologies in economic planning.

Forecasting Consumer Price Index in Personal Care Sector in Bukittinggi Using SVR with Grid Search and Radial Basis Function Kernel

2025-08-06T02:32:13+00:00

Inflation, measured by the Consumer Price Index (CPI), is vital for economic stability and policy making. In Bukittinggi, the Personal Care and Other Services sector shows notable CPI fluctuations, complicating accurate forecasting. This study uses Support Vector Regression (SVR) to predict monthly CPI data for this sector from 2020 to 2024. Data from Statistics Indonesia was normalized with Min-Max normalization to improve model accuracy and avoid scale distortion. Lag features were added to capture time dependencies, and data was split into training (80%) and testing (20%) sets. A linear SVR model was first applied but showed limited success due to the data’s non-linear nature. Therefore, the Radial Basis Function (RBF) kernel was used, with hyperparameters (C, sigma, epsilon, folds) optimized via Grid Search and cross-validation. The optimal settings (C=32, sigma=2, epsilon=0.1, k=10) yielded the lowest RMSE of 0.1099 in cross-validation and 0.0767 on testing. Results demonstrate that the RBF-SVR model effectively captures non-linear CPI patterns and outperforms the linear model. Evaluation metrics included RMSE, MSE, and MAE. The study concludes that SVR combined with Grid Search offers a robust forecasting method for sectors with complex CPI behavior, supporting local economic planning in Bukittinggi. Future research could investigate hybrid models and larger datasets to enhance prediction accuracy and adaptability to market changes.