Ertina Sabarita Barus
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ertina Sabarita Barus has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
Agung Prabowo
Program Study Sistem Informasi, Universitas Prima Indonesia, Medan, 20118, Indonesia
1 shared publicationD.A Barus
Program Study Fisika, FMIPA, Universitas Sumatera Utara, Medan, 20155 Indonesia
1 shared publicationDelima Sitanggang
Program Study Sistem Informasi, Universitas Prima Indonesia, Medan, 20118, Indonesia
1 shared publicationGellysa Urva
Program Study Teknik Informatika, Sekolah Tinggi Teknologi Dumai, Riau, 28815, Indonesia
1 shared publication- Muhammad Zarlis
Bina Nusantara University
1 shared publication Palma Juanda
Program Study Sistem Informasi, Universitas Prima Indonesia, Medan, 20118, Indonesia
1 shared publication
Latest Publications
- Open Access
- 29 July 2026
- Published by UIN Sumatera Utara Medan in ZERO: Jurnal Sains, Matematika dan Terapan
- Vol. 10 (2) , 671-682
Rule-Based IoT Decision Support System for Real-Time Chili Plant Growth Monitoring
This study proposes a rule-based IoT decision support system for real-time chili plant growth monitoring. Its main contribution is an integrated framework for chili cultivation that combines plant growth indicators and environmental sensor data within a single transparent IF-THEN rule structure an integration not commonly addressed in prior chili-monitoring systems. The system integrates eight parameters stem height, branches, leaves, fruits, soil moisture, temperature, water volume, and soil pH collected from 100 plant samples and 704 repeated environmental sensor records over a single 14-week growing season. Classification is formalized as a maximum-severity decision over a knowledge base of eight rules: each rule carries a severity level, and a record receives the label of the most severe triggered rule (Less Optimal > Fairly Good > Optimal). Thresholds were set through a two-step calibration that combines established agronomic ranges for Capsicum annuum with empirical refinement against observed plant responses. The logic was validated against 100 expert-labeled records using a confusion matrix, achieving 96% overall accuracy (95% CI: 90.2–98.4%), a macro-average F1-score of 93.4%, and per-class F1-scores between 88.9% and 98.0%. Classification results show that 75% of records fell into the Less Optimal category, driven primarily by elevated temperature (mean 34.09°C) and acidic soil pH (mean 5.38). The system delivers interpretable, actionable decision support without complex machine learning infrastructure; its main limitation is that thresholds were calibrated on a single site and season, so external multi-site validation is required before broader deployment.
- 29 July 2026
- Published by UIN Sumatera Utara Medan in ZERO: Jurnal Sains, Matematika dan Terapan
- Vol. 10 (2) , 671-682
Open AccessRule-Based IoT Decision Support System for Real-Time Chili Plant Growth Monitoring
This study proposes a rule-based IoT decision support system for real-time chili plant growth monitoring. Its main contribution is an integrated framework for chili cultivation that combines plant growth indicators and environmental sensor data within a single transparent IF-THEN rule structure an integration not commonly addressed in prior chili-monitoring systems. The system integrates eight parameters stem height, branches, leaves, fruits, soil moisture, temperature, water volume, and soil pH collected from 100 plant samples and 704 repeated environmental sensor records over a single 14-week growing season. Classification is formalized as a maximum-severity decision over a knowledge base of eight rules: each rule carries a severity level, and a record receives the label of the most severe triggered rule (Less Optimal > Fairly Good > Optimal). Thresholds were set through a two-step calibration that combines established agronomic ranges for Capsicum annuum with empirical refinement against observed plant responses. The logic was validated against 100 expert-labeled records using a confusion matrix, achieving 96% overall accuracy (95% CI: 90.2–98.4%), a macro-average F1-score of 93.4%, and per-class F1-scores between 88.9% and 98.0%. Classification results show that 75% of records fell into the Less Optimal category, driven primarily by elevated temperature (mean 34.09°C) and acidic soil pH (mean 5.38). The system delivers interpretable, actionable decision support without complex machine learning infrastructure; its main limitation is that thresholds were calibrated on a single site and season, so external multi-site validation is required before broader deployment.
- Open Access
- 30 June 2025
- Published by Private Company Technology Center in Eastern-European Journal of Enterprise Technologies
- Vol. 3 (2 (135)) , 42-53
Development of machine learning for forecasting optimization implemented in morphology plant growth
The object of the study is the forecasting and optimizing the plant growth rather. The data distribution at each iteration in the continuous optimization process tends to produce premature convergence because the optimum points are found at the beginning of the iteration, so that the actual optimum condition cannot be achieved. For this reason, a method is needed to see the optimum points at each iteration in the continuous optimization process. A multi-linear regression approach is used to predict the variables generated at each iteration, and then optimized using a neural network method approach for each optimum point found. This research is implemented and observed on the growth morphology of chili plants with a total sample of 100 stems, for 100 days of growth. The testing process consists of 5 different experimental scenarios based on the activation function, and the iteration process is carried out at 250, 500, and 1000 epochs. Furthermore, with a percentage of 70% training data and 30% testing data, the results obtained using the ReLU activation function have an ideal value compared to the Tanh, Softplus, Elu, and Sigmoid activation functions. Compared to the time series method with an MSE value of 4.62, this value is much better than the value of 8.6 for the time series. The RMSE and MAPE values of 16.36 and 36.53 are also excellent. Comparison of the level of forecasting accuracy of the results of continuous optimization carried out with the activation function ReLU and tanh compared to the time series method, the value with the activation function ReLU and tanh has a percentage value 46.36% and 46.86% and this value is a good value compared to using the time series method, which is exactly 67.39%
- 30 June 2025
- Published by Private Company Technology Center in Eastern-European Journal of Enterprise Technologies
- Vol. 3 (2 (135)) , 42-53
Open AccessDevelopment of machine learning for forecasting optimization implemented in morphology plant growth
The object of the study is the forecasting and optimizing the plant growth rather. The data distribution at each iteration in the continuous optimization process tends to produce premature convergence because the optimum points are found at the beginning of the iteration, so that the actual optimum condition cannot be achieved. For this reason, a method is needed to see the optimum points at each iteration in the continuous optimization process. A multi-linear regression approach is used to predict the variables generated at each iteration, and then optimized using a neural network method approach for each optimum point found. This research is implemented and observed on the growth morphology of chili plants with a total sample of 100 stems, for 100 days of growth. The testing process consists of 5 different experimental scenarios based on the activation function, and the iteration process is carried out at 250, 500, and 1000 epochs. Furthermore, with a percentage of 70% training data and 30% testing data, the results obtained using the ReLU activation function have an ideal value compared to the Tanh, Softplus, Elu, and Sigmoid activation functions. Compared to the time series method with an MSE value of 4.62, this value is much better than the value of 8.6 for the time series. The RMSE and MAPE values of 16.36 and 36.53 are also excellent. Comparison of the level of forecasting accuracy of the results of continuous optimization carried out with the activation function ReLU and tanh compared to the time series method, the value with the activation function ReLU and tanh has a percentage value 46.36% and 46.86% and this value is a good value compared to using the time series method, which is exactly 67.39%