Asep Saefuddin

The phenomenon of big data presents distinct challenges in the analysis process, especially when the data contains a very large number of variables. High complexity, potential redundancy, and the risk of overfitting are major issues that must be addressed through dimensionality reduction techniques. Principal Component Analysis (PCA) is a common method effective for data with linear relationships but has limitations in identifying nonlinear patterns. This research aims to improve performance of classification by introducing autoencoder for dealing with nonlinear relationship, data noise, missing values, outliers, and data with various scales. This study employs a quantitative approach through analysis of simulated and empirical data in the form of the Village Development Index from the Central Statistics Agency, which contains variables with various measurement scales. Both dimensionality reduction methods—PCA and neural network-based autoencoders—are tested across various data scenarios. The evaluation is conducted based on their effectiveness in preserving data structure, as well as the Mean Squared Error (MSE) values in the reconstruction process. The results indicate that PCA excels in computational efficiency and accuracy for data with linear relationships. In contrast, the autoencoder demonstrates superior performance in detecting nonlinear patterns, achieving lower Mean Squared Error (MSE) values with stable MSE standard deviations. Additionally, the autoencoder proves to be more robust in handling missing values and outliers compared to PCA. The selection of dimensionality reduction methods highly depends on the characteristics of the analyzed data. Autoencoders represent a superior alternative for handling complex and nonlinear data, although they require model parameter tuning. Further research is recommended to explore the influence of network architecture and training strategies of autoencoders on dimensionality reduction performance.

In the case of classification, model accuracy is expected to result in correct predictions. This study aims to analyze the performance of two kinds of support vector machine (SVM) methods: the support vector machine one versus one (SVM OvO) method and the generalized multiclass support vector machine (GenSVM) method. This method will compare to the generalized linear model, namely the multinomial logistic regression (MLR) method. Simulations were conducted using SVM OvO and GenSVM methods to get an overview of the parameters affecting both methods' performance. Furthermore, the three classification methods are implemented in the case of modelling the rice phenology and tested for performance. Simulation results show that, however, the SVM OvO and GenSVM machine learning methods are sensitive to the choice of model parameters. The empirical study results show that the SVM OvO and GenSVM methods can produce satisfactory model accuracy and are comparable to the MLR method. The best rice phenology model accuracy was obtained from the SVM OvO model, where 79.20 ± 0.21 overall accuracy and 70.69 ± 0.29 kappa were obtained. This research can be continued by handling samples, especially when class members are a minority, and can also add random effects to the SVM model.

Purpose: Classification of imbalanced data presents a major challenge in meteorological studies, particularly in rainfall classification where extreme events occur infrequently. This research addresses the issue by evaluating ensemble learning models in handling imbalanced rainfall data in Bogor Regency, aiming to improve classification performance and model reliability for hydrometeorological risk mitigation. Methods: Four ensemble methods: RF, RoF, DRF, and RoDRF were applied to rainfall classification using three resampling techniques: SMOTE, RUS, and SMOTE-RUS-NC. The data underwent preprocessing, stratified splitting, resampling, and 5-fold cross-validation. Performance was evaluated over 100 iterations using accuracy, precision, recall, and F1-score. Result: The combination of DRF with SMOTE-RUS-NC yielded the most balanced results between accuracy (0.989) and computation time (107.28 seconds), while RoDRF with SMOTE achieved the highest overall performance with an accuracy of 0.991 but required a longer computation time (149.30 seconds). Feature importance analysis identified average humidity, maximum temperature, and minimum temperature as the most influential predictors of extreme rainfall. Novelty: This research contributes a comprehensive comparison of ensemble forest-based methods for imbalanced rainfall data, revealing DRF-SMOTE as an optimal trade-off between performance and efficiency. The findings contribute to improved rainfall classification models and offer practical insight for disaster mitigation planning and resource management in tropical regions.

Purpose: This research aims to identify the optimal ensemble learning method for mitigating class imbalance in datasets utilizing various advanced techniques which include balanced random forest (BRF), SMOTE-random forest (SMOTE-RF), RUSBoost, and SMOTEBoost. The methods were systematically evaluated against conventional algorithms, including random forest and AdaBoost, across heterogeneous datasets with varying class imbalance ratios. Methods: This study utilized 13 secondary datasets from diverse sources, each with binary class outputs. The datasets exhibited varying degrees of class imbalance, offering scenarios to assess the effectiveness of ensemble learning techniques and traditional machine learning approaches in managing class imbalance issues. Study data were split into training (80%) and testing (20%), with stratified sampling applied to maintain consistent class proportions across both sets. Each method underwent hyperparameter optimization with distinct settings with repetition over 10 iterations. The optimal method was evaluated based on balanced accuracy, recall, and computation time. Result: Based on the evaluation, the BRF method exhibited the highest performance in balanced accuracy and recall when compared to SMOTE-RF, RUSBoost, SMOTEBoost, random forest, and AdaBoost. Conversely, the classical random forest method outperformed other techniques in terms of computational efficiency. Novelty: This study presents an innovative analysis of advanced ensemble learning techniques, including BRF, SMOTE-random forest, SMOTEBoost, and RUSBoost, which demonstrate significant effectiveness in addressing class imbalance across various datasets. By systematically optimizing hyperparameters and applying stratified sampling, this research produces findings that redefine the benchmarks of balanced accuracy, recall and computational efficiency in machine learning.

Peramalan deret waktu merupakan salah satu elemen kunci dalam analisis ekonomi dan keuangan. memungkinkan pemangku kepentingan untuk membuat perkiraan terhadap berbagai indikator ekonomi sebelum data resmi dirilis. Dalam konteks ini, model pembelajaran mesin seperti Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), dan Gated Recurrent Unit (GRU) menunjukkan potensi yang menjanjikan dalam memprediksi data deret waktu. Sejumlah penelitian juga menegaskan bahwa LSTM dan GRU mampu mengungguli kinerja RNN. Kedua model tersebut memiliki mekanisme untuk mengatasi masalah vanishing gradient yang sering ditemui pada model RNN konvensional. Penelitian ini menitikberatkan untuk menguji kinerja ketiga model tersebut pada data-data yang ada di Indonesia. Agar hasil lebih komprehensif, penelitian ini akan menguji model pada tiga jenis data yang berbeda meliputi IHSG, nilai ekspor dan PDB. Hasil penelitian ini mengindikasikan bahwa secara keseluruhan, model GRU menunjukkan kinerja terbaik, diikuti oleh model LSTM yang juga kompetitif dibandingkan RNN. Selain akurasi, model GRU dan LSTM juga menonjol dalam hal stabilitas kinerja, ditandai dengan simpangan baku yang relatif kecil jika dibandingkan dengan RNN. Keunggulan ini menjadi semakin signifikan terutama saat diterapkan pada model PDB dimana hanya tersedia untuk periode waktu yang pendek.

Indonesia faces severe challenges in addressing high levels of air pollution. To anticipate and address the effect of air pollution on human health and the environment, all parties need concerted efforts. Grouping regions based on similarities in air quality will help the government tailor policies more effectively. This research aims to conduct a cluster-based analysis of air quality in terms of not only its pollutant compound but also its spatial- and temporal context using tensor decomposition. The cluster elements were then analyzed with similar occurrence patterns to obtain potential factors. Air quality data containing time, region, and air quality parameters are processed to construct a tensor. Air quality parameters being addressed are carbon monoxide (CO), nitrogen dioxide (NO2), sulfur dioxide (SO2), ozone (O3), and particulate matter (PM10 and PM2.5). The data was retrieved from 31 representative monitoring stations across provinces in Indonesia from October 2023 to December 2023. Tensor decomposition with three latent features in each dimension was used for k-medoid clustering using the silhouette and elbow evaluation. Tucker decomposition is used for the general method in tensor decomposition because of its ability to control the decomposition rank. This means regulating the number of latent features or components to be extracted during the decomposition process. The analysis yielded optimal cluster numbers for region, time, and pollutant dimensions. It suggested that four clusters were identified based on region dimension. Meanwhile, in terms of temporal patterns, three clusters resulted. Furthermore, three clusters were identified based on pollutant types. The most concerning region cluster, including Palembang and Pontianak, have PM10 and PM2.5 concentrations above the safe level.

Multilevel Structural Equation Modeling (MSEM) is claimed to address hierarchical data structures and latent response variables, but it becomes unstable with an increasing number of levels. N-Level SEM (nSEM) is an SEM framework designed to handle a growing number of levels in the model. The nSEM framework uses the Maximum Likelihood Estimation (MLE) method for parameter estimation, which requires a large sample size and correct model specification. Therefore, it is essential to consider the necessary minimal sample size to ensure accurate and efficient parameter estimation in the nSEM model. This study examined how sample size affects the performance of parameter estimators in nSEM models. We propose a method to evaluate the effect of many environments to estimate the results of factor loadings and environmental variance produced by the model. In addition, we also assess the impact of environment size on the estimation results of factor loadings and individual variance. The results were then applied to actual data on student mathematics learning motivation in Depok. The findings show that neither the number of environments nor the size of the environment affects the performance of fixed parameter estimation in the nSEM model. nSEM indicates excellent performance in estimating environmental variance at level 2 when the number of environments increases. Conversely, increasing the size of the environment worsens the performance of estimating individual variance parameters. Overall, the nSEM framework for the latent random-intercept (LatenRI) model performs well with increasing sample sizes. The application data on LatenRI models show almost similar estimation results.Keywords: hierarchical data; latent random intercept model; multilevel structural equation modeling; n-level structural equation modeling.AbstrakMultilevel Structural Equation Modeling (MSEM) diklaim dapat mengatasi struktur data hierarki dan variabel respons laten, namun menjadi tidak stabil dengan bertambahnya jumlah level. N-Level SEM (nSEM) adalah kerangka kerja SEM yang dirancang untuk menangani semakin banyak level dalam model. Kerangka kerja nSEM menggunakan metode Maximum Likelihood Estimation (MLE) untuk estimasi parameter, yang memerlukan ukuran sampel yang besar dan spesifikasi model yang benar. Oleh karena itu, penting untuk mempertimbangkan ukuran sampel minimal yang diperlukan untuk memastikan estimasi parameter yang akurat dan efisien dalam model nSEM. Studi ini menguji bagaimana ukuran sampel mempengaruhi kinerja penduga parameter dalam model nSEM. Kami mengusulkan metode untuk mengevaluasi pengaruh banyak lingkungan dalam memperkirakan hasil factor loadings dan varians lingkungan yang dihasilkan oleh model. Selain itu, kami juga menilai dampak ukuran lingkungan terhadap hasil estimasi factor loadings dan varians individu. Hasilnya kemudian diterapkan pada data aktual motivasi belajar matematika siswa di Depok. Hasil menunjukkan bahwa baik jumlah lingkungan maupun ukuran lingkungan tidak mempengaruhi kinerja estimasi parameter tetap pada model nSEM. nSEM menunjukkan kinerja yang sangat baik dalam memperkirakan varians lingkungan pada level 2 ketika jumlah lingkungan meningkat. Sebaliknya, peningkatan ukuran lingkungan akan memperburuk kinerja pendugaan parameter varians individu. Secara keseluruhan, kerangka nSEM untuk model intersepsi acak laten (LatenRI) bekerja dengan baik dengan meningkatnya ukuran sampel. Data penerapan model LatenRI menunjukkan hasil estimasi yang hampir serupa.Kata Kunci: data hirarki; model intersep acak laten; model persamaan structural multilevel; model persamaan structural n-level. 2020MSC: 62D99

Sustainable Development Goals (SDGs) are developments that maintain sustainable improvement in society’s economic, social, and environmental welfare. Kemendes PDTT RI has issued the Village Development Index (VDI) to provide information and the status of village progress to support village development to improve the National SDGS. Modeling with multiple explanatory variables causes a high correlation between explanatory variables, multicollinearity, and coefficient estimation results, which have a large variance and overfitting in the prediction results. The modeling solution uses LASSO and GLASSO. The binary categorical response data use binary logistic regression (LR), so LR-LASSO and LR-GLASSO are used. North Maluku Province has a VDI ranking that tends to fall in 2018-2022. On the basis of the mean and variance of the coefficient estimation results and misclassification errors, LR-GLASSO is better than LR-LASSO and LR. LR-GLASSO is recommended for analyzing VDI data because it has many explanatory variables and the correlation between them is relatively high. The Indonesian government recommendation, if it is to increase the status of VDI in Indonesia, especially in the north Maluku province, is to increase the number of electricity users, food and beverage stores, and other cooperatives. The Indonesian government also needs to pay attention to villages relatively far from the regent's office, between food and beverage stalls, and supporting health centers, because they still need to be developed compared to other villages, and more than 50% of the villages are underdeveloped. If the Village SDGs are formulated by increasing the VDI status, it will support the achievement of the SDGs goals nationally.

Motivation to learn mathematics decreased due to the inability of teachers to implement innovative learning models and techniques. Therefore, this study aimed to investigate the effects of teachers' ability on students' motivation to learn mathematics by using quantitative methods and survey approaches. There were 32 mathematics teachers and 542 students in the 24 schools within the Depok region, selected as respondents through a stratified random sampling method. The research instruments of two questionnaires of teachers’ competence and students’ learning motivation were distributed to the respondents. Data analysis was conducted to test the random effect of teachers’ ability on students’ motivation to learn mathematics by using the effect of teachers’ random intercepts and competence as models 1 and 2, respectively. These two models were analyzed using the n-level Structural Equation Model (nSEM), and the result showed that model 2 was the best one to investigate the random effect of teachers’ ability and students’ learning motivation. The data analysis showed that the variance among teachers’ ability (0,0027) was less than learning motivation among students (0.0597). These findings indicated that the motivation levels of students taught by the same teacher varied significantly, whereas the effects of the teachers were relatively homogeneous. In other words, teachers’ ability was somewhat the same in increasing students’ learning motivation. Based on these findings, this research work suggests teachers keep improving their teaching techniques. Hence, students will be well motivated to learn so that the learning objectives will be well achieved.

Support Vector Machines (SVMs) have received extensive attention over the last decade because it is claimed to be able to produce models that are accurate and have good predictions in various situations. This study aims to test the SVR (Support Vector Regression) method for modeling the growth phase of paddy using sentinel-1 image data. This method was compared for its accuracy with the LR (Linear Model) method using RMSE and R2 statistics and model stability using 10 repetitions. The accuracy of the model with the two best predictors is when the NDPI and API Polarization Index are the predictors. The paddy age model from the SVR method is better than the paddy age model from the LR method, where the SVR method produces a model with an average RMSE of 11.13 and an average coefficient of determination of 88.10%. The accuracy of the SVR model with NDPI and API predictors can be improved by adding VH polarization to the model, where the average RMSE statistic decreases to 11.0 and the average coefficient of determination becomes 88.42%. In this scenario, the best model gives a minimum RMSE value of 10.35 and a coefficient of determination of 90.05%.