Journal Global Technology Computer

Optimizing Agricultural Commodity Price Forecasts Using an Ensemble Stacking Method Based on Market and Product Characteristics

2026-08-31T13:51:59+07:00

Commodity price prediction is a vital aspect of supporting decision-making because price fluctuations can create uncertainty for businesses and stakeholders. This study aims to compare the performance of several machine learning algorithms for commodity price prediction, namely Random Forest, XGBoost, Support Vector Regression (SVR), Gradient Boosting, and Stacking Ensemble. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE), and the coefficient of determination (R²). The experimental results show that Gradient Boosting achieved the best overall performance, with an RMSE of 219.07 and an R² of 0.9968, while Random Forest had the lowest MAE of 65.98. The Stacking Ensemble also demonstrated competitive performance, achieving an RMSE of 253.07 and an R² of 0.9958. In contrast, SVR produced the lowest performance, with an RMSE of 1,919.65 and an R² of 0.7555. Based on these results, Gradient Boosting was selected as the most optimal model because it provided the lowest prediction error and the highest ability to explain data variation. These findings demonstrate that selecting an appropriate machine learning algorithm plays a crucial role in improving commodity price prediction performance.

Penerapan Algoritma Apriori pada Asosiasi Generasi Z dan Domisili Terhadap Partisipasi Pemilih Muda pada Pemilu dan Pilkada 2024

2026-08-31T13:52:00+07:00

The 2024 General Election and Regional Election recorded a significant phenomenon in which young voters from Generation Z and millennials dominated the national permanent voter list, accounting for 56.45% of the total electorate. However, this high proportion was not necessarily followed by optimal participation rates, particularly at the regional level such as in Binjai City. This study aims to analyze the participation patterns of Generation Z in the 2024 General Election and Regional Election in Binjai City using the Apriori algorithm within the CRISP-DM framework. The dataset was obtained from the Binjai City KPU recapitulation, consisting of 1,201 Gen Z voter records with four attributes: voter age category, sub-district, general election attendance status, and regional election attendance status. The modeling process was applied with a minimum support of 0.04 and a minimum confidence of 0.5, generating 19 association rules with all lift values exceeding 1. The main findings indicate that Middle Gen Z (aged 20–23 years) is the most dominant group in attending the General Election, Binjai Barat Sub-district recorded the highest participation rate, and a selective participation pattern was identified in which non-attendance at the Regional Election does not necessarily reflect overall political apathy. The findings of this study are expected to serve as recommendations for the Binjai City KPU in designing more targeted strategies to improve young voter participation.

Penerapan Metode SMART pada Sistem Pendukung Keputusan untuk Menentukan Penerimaan Bantuan KIP Berdasarkan Kriteria Kelayakan Penerima

2026-09-28T21:25:24+07:00

Determining recipients of the Kartu Indonesia Pintar (KIP) educational assistance program is a process that requires accuracy in the selection process to ensure that assistance is provided to eligible candidates. The large number of applicants and the various criteria that need to be considered can make the assessment process complex when performed manually. This study aims to implement the Simple Multi Attribute Rating Technique (SMART) method in a Decision Support System (DSS) to assist in determining the priority of KIP assistance recipients in a more objective, structured, and measurable manner. The research method consists of determining candidate alternatives, establishing criteria and weights based on their level of importance, normalizing the weights, determining the utility value of each alternative, calculating the final score, and ranking the alternatives. Candidate data are used as alternatives and assessed based on predetermined eligibility criteria. The results of the SMART method calculation show that each alternative obtains a different final score according to its performance on each criterion. Based on the ranking results, alternative A12 obtained the highest score of 1.107, making it the alternative with the highest priority and recommended as a potential recipient of KIP assistance. The implementation of the SMART method can support a more systematic, transparent, and consistent selection process while providing ranking information that can be used as a consideration by decision makers. Therefore, the SMART method can be applied as an approach to support the determination of KIP assistance recipient priorities based on predetermined eligibility criteria.

Prediksi Kelulusan Tepat Waktu Mahasiswa Menggunakan Algoritma Random Forest Berdasarkan Data Akademik Mahasiswa

2026-09-29T08:27:26+07:00

Ontime graduation is one of the important indicators for measuring students’ success in completing their education within the prescribed study period. Higher education institutions face challenges in optimally utilizing historical academic data to identify students who are potentially at risk of delayed graduation at an early stage. This condition may limit the effectiveness of academic monitoring and the provision of appropriate assistance. This study aims to apply the Random Forest algorithm to predict students’ on-time graduation status based on academic characteristics and to evaluate the model’s ability to distinguish between students predicted to graduate on time and those predicted not to graduate on time. The dataset consists of 300 student records with attributes including Semester 1 to Semester 4 GPA, cumulative GPA, total credits, number of repeated courses, attendance percentage, and academic status. The research stages include data collection, preprocessing, dataset splitting into training and testing sets using an 80:20 proportion, Random Forest model development using 100 Decision Trees, classification, and model evaluation using a confusion matrix, accuracy, precision, recall, and F1-score. The contribution of this study is to provide a machine learning-based approach that utilizes historical academic data to generate predictive information regarding students’ graduation status and support more systematic academic monitoring. Testing on 60 records produced 29 True Positives, 23 True Negatives, 3 False Positives, and 5 False Negatives, with an accuracy of 86.67%, precision of 90.63%, recall of 85.29%, and F1-score of 87.88%. These results indicate that Random Forest is capable of classifying students’ graduation status with satisfactory performance on the dataset used.

Analisis Sentimen Masyarakat terhadap Kenaikan Harga Bahan Pokok Menggunakan Text Mining dan Pembobotan TF-IDF

2026-10-21T19:02:16+07:00

Rising prices for basic commodities represent an economic issue that can affect public purchasing power and trigger various reactions on social media. Given the vast volume of public opinion shared on social media, manual analysis is difficult; therefore, a text-processing approach is required to identify sentiment trends. This study aims to analyze public sentiment regarding the rising prices of basic commodities based on comments posted on Twitter, utilizing a text-mining approach and Term Frequency–Inverse Document Frequency (TF-IDF) weighting. Data were collected through a crawling process using keywords related to the increase in basic commodity prices. A total of 100 entries were gathered, from which 50 were selected for analysis. The processing stages included case folding, tokenizing, filtering, and stemming. Subsequently, specific terms assumed to represent positive or negative sentiments were identified to calculate their term frequency, document frequency, and TF-IDF weights. The results indicate that positive sentiment accounts for 24.241%, while negative sentiment accounts for 75.759%. The prevalence of terms categorized as negative suggests a negative trend within the analyzed data. This study is intended to provide an initial overview of public opinion regarding the rising prices of basic commodities and to serve as supporting information for further research or policy evaluation. However, these findings represent only the analyzed sample and cannot be generalized to the entire population. Furthermore, these percentages are derived from keyword-based calculations and TF-IDF weighting, rather than representing the accuracy of a classification model.