Journal of Soft Computing Exploration

Grape leaf disease classification using efficientnet feature extraction and catboostclassifier

Aditya Yoga Darmawan — 2025-03-17

Grapes are one of the most extensively cultivated crops worldwide due to their significant economic importance. However, the productivity of grape crops is often threatened by diseases caused by bacterial, fungal, or viral infections. Traditionally, the detection of infected grape leaves has been conducted through manual visual inspections, a method that is both time-consuming and prone to biases. Recent studies have leveraged transfer learning models to classify grape leaf diseases with high accuracy. Despite this progress, there is a notable gap in research exploring the integration of transfer learning for feature extraction and machine learning for feature classification in detecting grape leaf diseases. This study introduces a novel approach that combines transfer learning using EfficientNetB0 for feature extraction with a machine learning model, specifically Categorical Boosting (CatBoost), for feature classification. The proposed model demonstrates outstanding performance, achieving an accuracy of 99.56% on the test dataset, surpassing traditional transfer learning methods reported in previous studies.

Rainfall forecasting using triple exponential smoothing for rice cultivation in lamongan, jawa timur

Shafrila Widyantri — 2025-03-17

Rice cultivation is a major agricultural activity that is heavily influenced by weather conditions. Extreme weather events, such as heavy rainfall, can cause farmers' productivity to decline. Rainfall forecasts are important for farmers to help them make the right decisions in managing their farming businesses. This research aims to predict rainfall in Lamongan Regency, East Java province, and provide valuable information to rice farmers to plan the optimal planting season. The method used in this study is Triple Exponential Smoothing (TES), an effective forecasting technique for processing time series data with seasonal patterns. Monthly rainfall data for the last five years formed the basis of the forecast, with data sourced from NASA's Power Data Access Viewer. The analysis results include a Mean Absolute Percentage Error (MAPE) value of 97.559% for rainfall. This rainfall forecast can assist farmers in increasing rice productivity and minimizing the risk of crop failure due to unpredictable weather conditions. With the rainfall weather forecast, farmers are expected to know the suitable months for rice cultivation so that productivity increases

An advanced logistic regression model for forecasting payer revenue in private hospitals: a case study in manado

Wilsen Mokodaser — 2025-03-17

Manado, the provincial capital, stands as a vital center for healthcare services, where private hospitals compete intensively to attract patients from various economic and social backgrounds. Accurate revenue forecasting for partnered payers is essential for effective management strategies. This study employs a logistic regression model, achieving a notable accuracy of 79.55% in predicting hospital revenue based on payer partnerships. The confusion matrix reveals 21 true negatives (TN), confirming the model accurately identified low-revenue customers, with zero false positives (FP), indicating no misclassification of these individuals. However, 9 false negatives (FN) highlight a critical risk, as high-revenue customers were miscategorized as low revenue, even though 14 true positives (TP) were precisely identified. Based on these insights, hospitals can strategically target 61 payers projected to exceed median revenue, presenting a significant opportunity for income growth. Conversely, the 159 payers identified as below median revenue warrant urgent attention. To enhance engagement and increase revenue from these lower-revenue groups, targeted business strategies such as intensified marketing, personalized service offerings, and promotional discounts are recommended. This research contributes a novel approach to leveraging predictive analytics in healthcare, underscoring the pressing need for hospitals to innovate their operational strategies to optimize revenue in a competitive landscape.

Implementation of internet of things for leakage current monitoring system in medical equipment

Dio Alif Pradana — 2025-03-26

The rise in electricity consumption, especially in the health sector, has heightened concerns about electrical safety, particularly leakage current in medical equipment. The main objective of this research is to develop an IoT-based leakage current monitoring system specifically designed for low-voltage medical devices, aiming to enhance safety and prevent electrical hazards such as electric shocks and equipment damage. The system used two current sensors module (PZEMT-004T) to measure leakage at points near the voltage source and medical components. Data were processed by a microcontroller and transmitted to a web server for real-time monitoring via mobile devices. Testing on humidifiers and ECGs showed average accuracies of 90.11% and 92.49%, respectively, within a 10 mA range. However, the system could not detect currents below the 3 mA safety threshold because of the sensors reading limitation at 10 mA, indicating a need for sensor improvements. The IoT-based system enhances medical equipment safety, with future work focusing on better sensors and AI for predictive maintenance.

Improved human image density detection with comparison of YOLOv8 depth level architecture and drop-out implementation

Winda Yulita — 2025-03-29

Energy inefficiency due to Air Conditioners (AC) running in empty rooms contribute to unnecessary energy consumption and increased CO₂ emissions. This study explores how different depth levels of the YOLOv8 architecture and dropout regularization can enhance human density detection for smarter AC control systems. By evaluating model accuracy through Mean Average Precision (mAP50-95), we provide quantitative insights into how these modifications improve detection performance. Our dataset consists of 1363 images taken in an office environment at ITERA under varying lighting conditions and different human presence densities. The results show that the YOLOv8m model performs best, achieving an mAP50-95 score of 0.814 in training and 0.813 in validation, outperforming other YOLOv8 variants. Furthermore, applying dropout regularization improves model generalization, increasing mAP50-95 from 0.552 to 0.6 and effectively reducing overfitting. This study highlights the balance between architectural depth and dropout regularization in YOLOv8, demonstrating its effectiveness in energy-efficient smart buildings. The findings support the potential of deep learning-based human density detection in improving energy conservation strategies, making it a valuable solution for intelligent automation systems.

Optimizing Seq2Seq LSTM for Regional-to-National language translation on a web platform

Dwi Intan Af'idah — 2025-04-09

Machine translation for low-resource languages remains a significant challenge due to the lack of parallel corpora and optimized model configurations. This study developed and optimized a Seq2Seq Long Short-Term Memory (LSTM) model for Tegalan-to-Indonesian translation. A manually curated parallel corpus was constructed to train and evaluate the model. Various hyperparameter configurations were systematically tested, with the best-performing model achieving a BLEU score of 11.7381 using a dropout rate of 0.5, batch size of 64, learning rate of 0.01, and 70 training epochs. The results demonstrated that higher dropout rates, smaller batch sizes, and longer training durations enhanced model generalization and translation accuracy. The optimized model was deployed into a web-based application using Streamlit, ensuring accessibility for real-time translation. The findings highlighted the importance of hyperparameter tuning in neural machine translation for low-resource languages. Future research should explore Transformer-based architectures, larger datasets, and reinforcement learning techniques to further enhance translation quality and generalization.

Identification of lung cancer using gray level co-occurrence matrix (GLCM) and artificial neural network with backpropagation algorithm

Haniifah Hana Fauziah — 2025-04-25

Air pollution is a problem that occurs in various countries, including Indonesia. One of the consequences of poor air quality due to air pollution is health problems in the lungs, one of which is lung cancer. According to WHO data, lung cancer caused 1.80 million deaths in 2020. This is due to limited services to identify lung cancer early, resulting in delays in treatment. This study aims to identify lung cancer using CT-Scan image processing. The identification method uses a Backpropagation Artificial Neural Network (ANN BP) with Gray Level Co-occurrence Matrix (GLCM) feature extraction. Preprocessing is carried out to improve image quality by removing noise using a median filter. Segmentation of preprocessing results using Otsu threshold. Texture features from segmentation can be calculated from the resulting GLCM, such as Angular Second Moment (ASM)/energy, contrast, correlation, Inverse Different Moment (IDM)/homogeneity, and entropy. These values are obtained from angles of 0°, 45°, 90°, and 135°, and a distance between pixels of 2 pixels. Identification using ANN with Backpropagation algorithm. This study used images of normal lungs and lung cancer with 160 training data images and 40 test data images. The best test results were obtained with the best accuracy level of 92.5%.