cover
Contact Name
Muh. Isbar Pratama
Contact Email
isbarpratama@unm.ac.id
Phone
+6285399692435
Journal Mail Official
jmathcos@unm.ac.id
Editorial Address
Kampus Parangtambung UNM, Jl. Dg. Tata Raya Prodi Matematika Lt. 3 Gd FG Jurusan Matematika FMIPA
Location
Kota makassar,
Sulawesi selatan
INDONESIA
Journal of Mathematics, Computation and Statistics (JMATHCOS)
ISSN : 24769487     EISSN : 27210863     DOI : https://doi.org/10.35580/jmathcos
Core Subject : Education,
Fokus yang didasarkan tidak hanya untuk penelitian dan juga teori-teori pengetahuan yang tidak menerbitkan plagiarism. Ruang lingkup jurnal ini adalah teori matematika, matematika terapan, program perhitungan, perhitungan matematika, statistik, dan statistik matematika.
Articles 269 Documents
ARIMA Modeling With Expectation–Maximization Kalman Imputation for Missing Values in Gorontalo Red Chili Prices Moh Fiqri Nurwahidin; Isran K. Hasan; Muhammad Rezky Friesta Payu
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/kfasnm79

Abstract

The price of red chili peppers at Gorontalo City Central Market fluctuates considerably and is classified as a volatile food commodity that can contribute to regional inflation. Accurate forecasting is therefore important for supporting price monitoring and planning. However, daily price records contain missing values, particularly on holidays when price recording was not conducted. This study applies the Expectation-Maximization Kalman (EM-Kalman) algorithm to impute missing observations before modeling the price series using ARIMA. The analysis uses daily price data from January 2023 to December 2025, consisting of 1,096 observations, of which 355 (32.39%) were missing. The data were divided into 80% training and 20% testing sets. The EM-Kalman algorithm successfully produced a continuous time series and converged at the 98th iteration. The ARIMA(0,1,2) model was selected based on the smallest AIC and BIC values, with statistically significant MA parameters and residuals satisfying the white-noise assumption based on the Ljung-Box test. The resulting model achieved a MAPE of 7.56% for the training data and 4.96% for the testing data, indicating highly accurate forecasting performance based on the adopted MAPE classification. These results indicate that the EM-Kalman and ARIMA combination can be used to reconstruct the incomplete price series and generate accurate forecasts of red chili pepper prices. However, this study does not compare EM-Kalman with alternative imputation methods or ARIMA with benchmark forecasting models, which should be considered when interpreting the forecasting performance.
Hybrid Spatio-Temporal Modeling of Dengue Incidence Using ST-CAR-ARIMA with Remote Sensing Covariates in Makassar, Indonesia BQ Nurkhalishah K.; Aswi Aswi; Suwardi Annas; Rahmat Hidayat; Nur Salsabila Wero; Sukarna Sukarna
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/yyrha036

Abstract

Dengue hemorrhagic fever (DHF) exhibits complex spatial and temporal dynamics that require analytical approaches capable of jointly accounting for geographic heterogeneity, temporal dependence, and environmental conditions. This study developed a hybrid spatio-temporal framework integrating a Bayesian Spatio-Temporal Conditional Autoregressive (ST-CAR) model with a Seasonal Autoregressive Integrated Moving Average (SARIMA) model to characterize and predict DHF incidence in Makassar City, Indonesia. Monthly DHF cases from 149 urban villages were analyzed over 32 months, from January 2022 to August 2024, together with remotely sensed environmental covariates represented by the Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI). A Bayesian ST-CAR model with a Poisson likelihood was first fitted to account for spatial heterogeneity, temporal dependence, and environmental effects. The model revealed substantial residual spatial heterogeneity and temporal dependence, whereas spatial dependence was relatively weak. NDVI and NDWI showed negative posterior mean associations with DHF incidence; however, their 95% credible intervals included zero. Residual temporal analysis was subsequently performed to determine whether additional temporal structure remained after ST-CAR estimation. Among the candidate seasonal models, SARIMA (0,0,0) (1,0,0)6 produced the lowest AIC, AICc and BIC, indicating residual temporal dependence at a six-month seasonal lag. Incorporating this residual component into the ST-CAR predictions reduced the RMSE from 1.533081 to 1.533053, corresponding to an improvement of only approximately 0.0018%. Thus, the hybrid model provided negligible improvement in overall point-prediction accuracy, suggesting that the ST-CAR component had already captured most of the systematic spatio-temporal variation in DHF incidence. Nevertheless, the SARIMA component provided additional characterization of residual six-month temporal dependence. Overall, the proposed framework demonstrates the utility of combining Bayesian spatio-temporal modeling, remote-sensing covariates, and residual seasonal time-series analysis to distinguish dominant spatio-temporal variation from secondary residual temporal structure in dengue incidence.
Comparative Performance of Backpropagation and Radial Basis Function for Cum Laude Classification Nurkhalifah Anwar; Suwardi Annas; Rahmat Syam
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/ff721j16

Abstract

Higher-education institutions store large academic records, but these data often remain administrative and provide limited support for model-based evaluation of graduation outcomes. In neural-network classification, method selection is important because Backpropagation and Radial Basis Function (RBF) networks learn through different mechanisms: Backpropagation iteratively adjusts weights by propagating error, whereas RBF represents data through distance-based responses around selected centres. Consequently, their accuracy, error balance, and computational time cannot be assumed to be equivalent for educational datasets and need to be compared under the same testing conditions. The comparison was intended to support method selection for the available alumni dataset rather than to propose an institution-wide deployment model. This study compared Backpropagation and RBF neural networks for retrospective classification of undergraduate alumni into Cum Laude and Non-Cum Laude groups and examined the predictive importance of the input variables. The dataset consisted of 824 alumni records from the Department of Mathematics, Universitas Negeri Makassar, with eight predictors: study programme, sex, admission route, study duration, tuition fee, graduated credits, age at entry, and final grade point average (GPA). After data cleaning, encoding, and Min-Max normalisation, the data were evaluated using a stratified 80:20 split, resulting in 659 training records and 165 test records. Twelve Backpropagation scenarios and twelve RBF scenarios were tested. Under the retained split and parameter grid, the best Backpropagation model used an 8-8-1 architecture and a learning rate of 0.01, achieving 98.79% accuracy and a 98.41% F1-score. The best RBF model used 40 centres and a spread of 0.8, achieving 87.88% accuracy and an 85.71% F1-score. RBF trained faster in the recorded Google Colab run, but Backpropagation produced fewer classification errors. Permutation importance showed that study duration (55.84%) and final GPA (35.17%) dominated the selected model. Because both variables are observed near graduation and are closely related to the administrative criteria for Cum Laude, the selected model should be interpreted as retrospective classification rather than early prediction.
Implementation of Robust Spatial Autoregressive Model (RSAR) in Modeling Toddler Stunting in Indonesia Dewi Aprilia Wardani Syam; Wahidah Sanusi; Ruliana Ruliana
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/2vbxp780

Abstract

This study employed the Spatial Autoregressive (SAR) model and the Robust Spatial Autoregressive (RSAR) model to analyse the prevalence of stunting among children under five in Indonesia, accounting for both spatial dependence among provinces and the presence of spatial outliers. The study utilised secondary data for 2024 covering all 38 provinces of Indonesia. Stunting prevalence was used as the response variable, while eight explanatory variables were considered: exclusive breastfeeding coverage, complete basic immunisation coverage, access to adequate sanitation, access to clean water, the proportion of the population living below the poverty line, adequate housing, early marriage prevalence, and family planning participation. The analysis began with the construction of a spatial weight matrix using the K-Nearest Neighbour (KNN) approach, followed by the assessment of spatial autocorrelation using Moran’s I. Subsequently, Ordinary Least Squares (OLS) regression was performed, followed by spatial effect testing using the Lagrange Multiplier test, SAR modelling, detection of spatial outliers through Moran’s scatterplot analysis, and RSAR modelling using the MM-estimation approach. The results revealed significant positive spatial autocorrelation in stunting prevalence across Indonesian provinces, indicating that stunting conditions in a province are associated with those of neighbouring provinces. The SAR model identified exclusive breastfeeding coverage, complete basic immunisation coverage, the proportion of the population living below the poverty line, and early marriage prevalence as significant determinants of stunting prevalence. In contrast, the RSAR model identified five significant factors: exclusive breastfeeding coverage, complete basic immunisation coverage, access to adequate sanitation, the proportion of the population living below the poverty line, and early marriage prevalence. The significance of adequate sanitation emerged after the influence of spatial outliers had been reduced through robust estimation. In terms of model performance, the RSAR model achieved an AIC value of 130.41. Furthermore, the residuals of the RSAR model exhibited no significant spatial autocorrelation, as indicated by a Moran’s I value of 0.0551 (p = 0.1202). These findings suggest that the RSAR model provides a more reliable representation of the factors influencing stunting prevalence in Indonesia, as it effectively accounts for spatial dependence while reducing the impact of spatial outliers.
Deepfake Audio Detection Using Hybrid Audio Spectrogram Transformer with Dual-Attention Feature Fusion and Bayesian Optimization Kusnaeni Kusnaeni; Muhaimin Hading; Zaitun Zaitun; Andi Oxy Raihan Machikami R; Afif Budi Andy B
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/ja96za26

Abstract

Deepfake audio has become a serious threat to digital security due to its ability to generate highly realistic synthetic speech for malicious purposes such as impersonation, fraud, and social engineering. Despite the rapid advancement of deepfake generation technologies, audio-based detection remains relatively underexplored compared to image and video domains. Moreover, existing approaches largely rely on handcrafted features and conventional machine learning methods, which are often insufficient to capture complex temporal and spectral dependencies and often exhibit limited generalization to unseen deepfake generation techniques. This study proposes a novel framework that leverages log-Mel spectrogram representations to capture informative time–frequency characteristics of audio signals. The proposed model integrates a Hybrid Audio Spectrogram Transformer to model global dependencies through self-attention and is further enhanced by a Dual-Attention Feature Fusion mechanism to improve discriminative capability at both spatial and channel levels. In addition, Bayesian Optimization is employed to efficiently optimize hyperparameters, leading to improved model performance, faster convergence, and better generalization. Experimental results demonstrate that the proposed model achieves an accuracy of 93.60% and an AUC of 0.9724, indicating excellent discriminative performance. The model also achieves a high recall in detecting deepfake audio, highlighting its effectiveness in minimizing false negatives and improving robustness in real-world scenarios. These findings demonstrate that the proposed framework provides an effective and reliable solution for deepfake audio detection and offers significant potential for deployment in practical cybersecurity and digital audio forensic applications.
Block-Based K-Medoids Clustering For Profiling Regions in Central Java Based On Employment Indicators Rio Aristiarto; Kariyam Kariyam
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/hw6z1b10

Abstract

Employment is an important dimension of regional development because it reflects job opportunities, labour market participation, and social welfare. Single indicators cannot fully capture employment conditions, especially in regions with significant labour market disparities such as Central Java. This study groups regencies and cities in Central Java using multiple employment indicators and the Block-Based K-Medoids clustering algorithm. The study uses 2024 secondary data from the Central Java Statistics Agency, comprising the open unemployment rate, labour force participation rate, average monthly wage, proportion of youth aged 15–24 classified as NEET (not in education, employment or training), and number of industries. All variables were normalized with Min-Max normalization before clustering to ensure comparability across indicators with different measurement scales. The optimal number of clusters was determined using the Silhouette, Calinski-Harabasz, and Hartigan indices, all of which indicated that three clusters were optimal. The results show that Cluster 1 is characterized by relatively high unemployment and NEET rates and low wages, Cluster 2 represents regions with the most favourable employment conditions, and Cluster 3 has the highest average wages but the lowest labour force participation rate. These findings provide a useful basis for more targeted employment policy formulation in accordance with regional labour market characteristics, and they may support local governments in identifying priority areas for intervention and development planning.
Comparative Modelling of Sharia-Compliant Stock Returns and Volatility Using GRU–GARCH and DES–GARCH: Evidence from Telkom Indonesia Indrawan Indrawan; Fadeli Muhammad Habibie; Magdalena Effendi; Nurul Hidayat; Dewi Leonita Sari; Jesselyn Nixie Amadea; Nabela Ariyaninda Budi Astutik
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/5yr4h156

Abstract

The integration of deep learning and econometric models has become an active research direction in modern computational finance, particularly for capturing the nonlinear dynamics and time-varying volatility that characterize equity markets. In the Islamic capital market, reliable modeling of return and volatility matters for risk assessment and for investment decisions that stay consistent with Sharia principles. This study compares two hybrid models, GRU–GARCH and DES–GARCH, for jointly modeling the return and conditional volatility of PT Telekomunikasi Indonesia Tbk (TLKM), a highly liquid Sharia-compliant stock. A Gated Recurrent Unit (GRU) network and Double Exponential Smoothing (DES) were each used to model the conditional mean of the daily log returns, while the residuals of both were modeled with a GARCH specification to capture volatility clustering. The analysis drew on 2,463 daily observations from March 2016 to March 2026, split into 80% training and 20% testing data, with model selection based on validation performance and information criteria. Forecast accuracy was measured with the scale-independent Mean Absolute Scaled Error (MASE), complemented by the QLIKE loss for the variance forecasts, the Diebold–Mariano test for pairwise comparison, and a Kupiec backtest for Value-at-Risk. Both hybrids beat a naïve benchmark, and a GARCH(1,1) model with Student-t innovations adequately described the strong volatility persistence (about 0.98) and the heavy tails found in the return series. Contrary to the common expectation that deep learning improves accuracy, the parsimonious DES–GARCH model produced a lower MASE (0.914 against 0.983) and a lower QLIKE (2.68 against 2.83) than GRU–GARCH, and the gap in mean-forecast accuracy was statistically significant under the Diebold–Mariano test. DES–GARCH also delivered better-calibrated Value-at-Risk, passing the Kupiec test at both confidence levels while GRU–GARCH did not. These results suggest that, for daily Sharia-stock returns whose predictable component is small, a simpler smoothing-based mean model can match or exceed a deep learning model, and that a strong classical baseline should be established before more complex architectures are adopted for operational risk work.
Comparative Evaluation of Long Short-Term Memory and Support Vector Regression for Daily Rice Price Forecasting in Makassar Arwini Arisandi; Ismail Gaffar; Sitti Nurhaliza; Hardianti Hafid
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/pphv6749

Abstract

Rice is a strategic food commodity that plays an important role in food security, household purchasing power, and regional economic stability. Fluctuations in daily rice prices can affect inflation, market behavior, and policy decisions related to food distribution and price stabilization. This study aims to comparatively evaluate the performance of Long Short-Term Memory (LSTM) and Support Vector Regression (SVR) models in forecasting daily rice prices, particularly for Premium Rice and Medium Rice. The study initially used 1,077 observations of daily rice price time-series data. Following preprocessing, including the construction of lag and moving-average features, 1,047 observations remained for modeling. Descriptive analysis showed that Premium Rice had a minimum price of IDR 11,000, a maximum price of IDR 16,250, a mean price of IDR 13,690.55, and a standard deviation of 1,300.812. Meanwhile, Medium Rice had a minimum price of IDR 9,800, a maximum price of IDR 14,760, a mean price of IDR 12,212.12, and a standard deviation of 1,334.867. The LSTM model was implemented using Python, whereas the SVR model was implemented using R. Both models were evaluated using the same chronological data partitions and testing period. Model performance was assessed using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). Both models demonstrated accurate forecasting performance, with MAPE values below 10%. Under the experimental configuration and common testing period used in this study, SVR achieved lower forecasting errors than LSTM for both rice categories. For Premium Rice, SVR produced an MAE of 61.18, an RMSE of 159.78, and a MAPE of 0.42%, whereas LSTM produced an MAE of 180.00, an RMSE of 293.00, and a MAPE of 1.18%. For Medium Rice, SVR produced an MAE of 67.03, an RMSE of 209.29, and a MAPE of 0.49%, compared with LSTM, which produced an MAE of 181.00, an RMSE of 284.00, and a MAPE of 1.32%. These findings indicate that SVR was more suitable for the data and feature configuration applied in this study.
A Fuzzy K-Prototypes Ensemble SMOTE-NC SVM Framework for Open Unemployment Classification Setyo Wahyudi; Indahwati Indahwati; Made Sumertajaya
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/39jhr427

Abstract

Open unemployment classification using labor force survey data presents a significant challenge because data are highly imbalanced and consist of both numerical and categorical variables. Conventional classification models often achieve high overall accuracy but exhibit limited capability in identifying unemployed individuals, who represent the minority class. This study proposes a hybrid classification framework Fuzzy K-Prototype clustering, Synthetic Minority Over-sampling Technique for Nominal and Continuous variables (SMOTE-NC), support vector Machine, and weighted ensemble learning to improve minority-class identification in mixed-type labor force survey data. Unlike previous clustering-based ensemble approaches that primarily employ K-Means and conventional SMOTE, the proposed framework utilized Fuzzy K-Prototypes to accommodate mixed-type variables and SMOTE-NC to generate synthetic minority observations within homogenous clusters. The model was evaluated using the 2024 National Labor Survey (Sakernas) data from West Java Province, Indonesia. Experimental result indicate that proposed framework increased Recall (Sensitivity) from 73.93% to 77.94%, demonstrating an improved ability to identify unemployed individuals. Although Accuracy, Precision, Specificity, F1-score, and Balanced Accuracy decreased compared with the conventional SVM model, the proposed framework provides a more appropriate classification strategy for highly imbalance labor force datasets, where accurate identification of unemployed individuals is more important than maximizing overall classification accuracy, thereby offering valuable support for labor market analysis and employment policy development.
Integrated Force Field Analysis (FFA) and Inferential Statistics Method for Student Recruitment Strategy Evaluation Wiwik Wiyanti; Jacob Stevy Seleky; Rambu Marshanda Tamu Ina Paa
Journal of Mathematics, Computations and Statistics Vol. 9 No. 3 (2026): Volume 09 Issue 03 (September 2026)
Publisher : Jurusan Matematika FMIPA UNM

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35580/ebwk5a42

Abstract

This study applies an integrated approach combining inferential statistics, specifically inferences for two population means, with the Force Field Analysis (FFA) method. This study aims to analyze marketing strategies to increase new student enrollment at a university in Banten Province, Indonesia. The strategies analyzed in this study were classified into two categories: driving forces and restraining forces. The subjects of this study were employees from the Sales and Marketing Division of the university. The data analyzed in this study consisted of primary data collected through the administration of a structured questionnaire. The data analyzed in this study were primary data collected through a questionnaire. The questionnaire responses were categorized into two groups: driving forces and restraining forces. Data were analyzed using an integrated approach that combined FFA with inferential statistics, specifically the independent-samples t-test. Prior to its administration to the research subjects, the questionnaire underwent an expert validation process to ensure the relevance and appropriateness of its items. Descriptively, the mean score for the driving forces was 8.3, which was higher than that of the restraining forces was 5.4. The t-test yielded a calculated  ( ) of 11.4, which was greater than the critical t-value ( ) of 1.67. It’s means  that marketing strategies classified as driving forces had significantly higher mean scores than those classified as restraining forces in efforts to recruit prospective new students to the university over the past three years. Furthermore, the findings revealed two restraining forces with higher mean scores than the driving forces, specifically for Items 32 and 33, which concerned heavy workload and an inadequate compensation and incentive system, respectively.