Bulletin of Information Technology (BIT)

Prediksi Penjualan Menggunakan Exponential Smoothing sebagai Pendukung Business Intelligence UMKM Kuliner

2026-09-23T23:23:38+00:00

Micro, Small, and Medium Enterprises (MSMEs) in the culinary sector face challenges in sales management due to unpredictable demand fluctuations. This study aims to apply the Exponential Smoothing method as part of a time series approach to predict daily sales of fried chicken and catfish pecel, and integrate the results into a Business Intelligence (BI) framework to support operational and strategic decision-making. Primary data were collected through unstructured interviews with the business owner during Januari-December 2025, with a total of 313 operating days. Modeling used alpha (α) = 0.3 selected through trial and error to obtain the best accuracy. Model evaluation used MAPE, MAE, and RMSE metrics. Results showed MAPE values of 19.33% for fried chicken (good accuracy) and 21.12% for catfish pecel (reasonably acceptable). MAE values were 3.58 and 3.25 portions/day respectively, while RMSE values were 4.42 and 4.09. Sales forecasts for January 2026 projected 23 portions/day for fried chicken and 19 portions/day for catfish pecel. Prediction results were visualized in a Business Intelligence dashboard using Microsoft Excel, producing data-driven managerial recommendations for production planning and inventory management.

Klasifikasi Jenis Ayam Menggunakan Ensemble Voting CNN (Mobilenetv2 Dan Efficientnet-B0) Berbasis Citra

2026-09-24T00:06:19+00:00

Digital image-based chicken species classification plays a crucial role in supporting the identification process in the livestock sector. However, the similarity in visual characteristics between chicken species makes the manual classification process inefficient and error-prone. This study aims to develop a chicken species classification model using the Ensemble Soft Voting method by combining two Convolutional Neural Network (CNN) architectures, namely MobileNetV2 and EfficientNet-B0. The dataset used consists of 1,500 images from five chicken classes: Bangkok Chicken, Broiler Chicken, Free-Range Chicken, Bantam Chicken, and Laying Hen. The research stages include preprocessing, data augmentation, transfer learning, model training, and combining prediction probabilities using the Soft Voting method. Evaluation is carried out using Accuracy, Precision, Recall, F1-Score, and Confusion Matrix metrics. The results showed that MobileNetV2 achieved an accuracy of 97.18%, EfficientNet-B0 achieved 96.48%, while the Ensemble Soft Voting method produced the best performance with an accuracy of 97.89%, a precision of 97.96%, a recall of 97.89%, and an F1-score of 97.88%. These results indicate that the Ensemble Soft Voting method is able to improve classification accuracy and stability compared to using a single model, so it has the potential to be applied as a solution for digital image-based chicken species identification.

Explainable Gradient Boosting for Egg Production Anomaly Detection in Laying Hen Farms

2026-09-23T23:23:39+00:00

Abstract- Small-scale laying-hen farms require analytical tools that can detect abnormal egg-production patterns while remaining interpretable for operational decision support. This study develops and evaluates an explainable gradient boosting framework for low-production anomaly detection using farm-level environmental, management, and production records. The dataset contained 481 observations with eight variables, including flock size, feed amount, ammonia, temperature, humidity, light intensity, noise, and total egg production. Exact-duplicate inspection found 407 duplicated rows; therefore, the 74 unique observations were used as the primary evaluation dataset. Egg production rate was derived by normalizing total egg output by flock size, and low-production anomalies were defined using the 10th percentile threshold, resulting in eight anomaly cases. Supervised tree-based models were evaluated using repeated stratified four-fold cross-validation with 20 repeats, and unsupervised methods were used as anomaly baselines. XGBoost achieved the best full-feature performance, with F1-score of 0.928, recall of 1.000, ROC-AUC of 0.997, and PR-AUC of 0.984. Sensitivity analysis without flock size showed that environmental and management variables retained useful signal. SHAP analysis indicated that flock size, light intensity, and temperature were the dominant predictors. The findings support interpretable operational monitoring of low egg-production patterns, although the results require validation on larger expert-labeled datasets.

Pengembangan Smart Egg Incubator Semi Otomatis Berbasis Internet of Things dengan Notifikasi Cerdas

2026-09-24T00:08:00+00:00

Abstract—Conventional egg incubation still faces challenges such as unstable temperature and humidity, manual egg turning, and limitations in continuously monitoring incubator conditions. These conditions can increase the monitoring burden and affect the stability of the incubation process. This study aims to develop a semi-automatic Smart Egg Incubator based on the Internet of Things (IoT) to perform real-time temperature and humidity monitoring, automatic egg turning, and a smart notification system for changes in incubation conditions. The research method employs a Research and Development approach with a prototyping methodology, including needs assessment, system design, hardware and software implementation, and functional testing using black-box testing. The system was developed using an ESP32, a DHT22 sensor, a heater, a DC fan, a servo motor, and an IoT platform as the data communication medium. Test results showed that the monitoring, temperature control, automatic egg turning, data communication, and notification functions operated as designed in the conducted test scenarios. The contribution of this research lies in the integration of real-time incubation condition monitoring, automatic egg turning, and smart notifications into a single semi-automatic IoT-based prototype. This integration provides a more practical and responsive alternative to manual monitoring and supports the development of more efficient egg-hatching systems.
Keywords: Internet of Things; Smart Egg Incubator; ESP32; Real-Time Monitoring; Automatic Egg Turner; Notification System.

Translated with DeepL.com (free version)

Klasifikasi Kinerja Konten YouTube Mobile Legends Menggunakan LightGBM dengan Pelabelan Berbasis K-Means dan Interpretasi SHAP

2026-09-24T07:56:19+00:00

YouTube is a video-sharing platform widely used by content creators, including those in the gaming category such as Mobile Legends. The intense competition among Mobile Legends content creators has made content performance analysis essential for developing data-driven content strategies. However, most previous studies have relied on manual data labeling, which is prone to subjectivity and lacks efficiency. Furthermore, high-accuracy classification models are often treated as black-box models, making it difficult to explain the factors influencing prediction outcomes. This study aims to develop a YouTube Mobile Legends content performance classification model by integrating K-Means, LightGBM, and Shapley Additive exPlanations (SHAP). The dataset consists of 3,052 videos collected using the YouTube Data API v3. The K-Means algorithm was employed to generate automatic labels, while LightGBM was used as the classification model and SHAP was applied to interpret the model's predictions. The results indicate that the optimal number of clusters was K=2, producing two performance classes: High Performance (72.44%) and Low Performance (27.56%). The dataset was subsequently divided into 80% for training and 20% for testing. On the testing dataset, the LightGBM model achieved an accuracy of 93.78%, precision of 93.87%, recall of 93.78%, and an F1-score of 93.81%. Furthermore, 10-fold cross-validation yielded a mean accuracy of 92.75% with a standard deviation of 0.0148, demonstrating the model's robustness and stability. SHAP analysis revealed that Subscriber Count was the most influential feature, followed by Duration, Video Age, and Title Length. These findings demonstrate that the integration of K-Means, LightGBM, and SHAP provides an effective approach for objectively, accurately, and transparently classifying the performance of Mobile Legends content on YouTube, while offering valuable insights to support content creators in developing data-driven content strategies.

Analysis of the Spatial Distribution of Rice Prices Acrossprovinces in Indonesia Using the K-Means Algorithm

2026-09-26T03:51:54+00:00

Differences in rice prices across provinces in Indonesia reflect variations in distribution and market conditions that can be analyzed using a data clustering approach. This study aims to map the characteristics of interprovincial rice prices using the K-Means Clustering algorithm. The dataset consists of prices for Premium Rice, Medium Rice, Non-SPHP Medium Rice, and SPHP Rice obtained from official government data. The research stages included data preprocessing, transformation, normalization using StandardScaler, determining the number of clusters using the Elbow Method and Silhouette Score, and visualizing the results using Principal Component Analysis (PCA). The Elbow Method results indicated an efficient point at K = 5, while the highest Silhouette Score of 0.5658 was obtained at K = 7; thus, seven clusters were selected as the best model. PCA visualization reveals that provinces in Indonesia exhibit rice price characteristics that form seven groups with varying degrees of similarity. The research results indicate that the K-Means algorithm is capable of effectively identifying rice price clustering patterns and producing a map that can support the analysis of inter-regional price characteristics as a basis for formulating food distribution policies and stabilizing rice prices in Indonesia.

Explainable Machine Learning for Post-Disaster Stunting Risk Prediction to Support Emergency Nutrition Decision-Making

2026-09-26T04:15:27+00:00

Stunting in children under five is a chronic nutritional disorder whose risk increases significantly in post-disaster settings due to disruptions in food supply chains, damage to health infrastructure, and deteriorating sanitation conditions. Delays in identifying children at high risk of stunting in evacuation centers prevent targeted, timely, and efficient emergency nutrition interventions. This study proposes a post-disaster stunting risk prediction system based on Prepared Machine Learning (XML), which integrates a predictive model based on ensemble learning with the SHapley Additive exPlanations (SHAP) interpretability framework to support transparent and evidence-based emergency nutrition intervention decision-making. The model is built with Explainable Artificial Intelligence (XAI) to make the stunting prediction system transparent and understandable. SHAP is used as the primary method to explain the contribution of each factor to stunting risk prediction. Data were processed using anthropometric and sociodemographic datasets of children aged 0–59 months, consisting of victim data collected from the three provinces of Aceh, North Sumatra, and West Sumatra, data from the Center for Disaster Management (CDC), and data and information from the Indonesian National Disaster Management Agency (BNPB). Twelve input features were used, including height-for-age (HAZ) z-score, weight-for-age (WAZ) z-score, exclusive breastfeeding, frequency of food intake, immunization status, household sanitation conditions, family income level, and duration of displacement. Model performance was assessed using Area Under the Receiver Operating Characteristic Curve (AUC-ROC), F1-Score, precision, and recall, with stratified tenfold cross-validation to ensure model generalizability. SHAP-based interpretability analysis identified the five features with the highest predictive contribution as HAZ z-score, duration of displacement, frequency of food intake, exclusive breastfeeding, and sanitation conditions, which consistently align with established stunting risk factors in the nutrition science literature. The system then interfaces into a web-based decision support interface that displays SHAP power plots per child and population-level least squares plots, enabling field nutrition workers without a data science background to understand the rationale behind each risk prediction. This study provides an explainable and operationally applicable machine learning framework for the Ministry of Health, BNPB, and National humanitarian agencies as a transparent, auditable, and accountable decision-support instrument for emergency nutrition interventions in post-disaster situations.