Journal of Research in Artificial Intelligence for Systems and Applications

CLASSIFICATION OF DISEASES ON MANGGO LEAVES USING ARCHITECTURE VGG16 AND SUPPORT VECTOR MACHINE (SVM)

Achmad Muchdor Firdaus, Dwiretno Istiyadi Swasono, Januar Adi Putra — Wed, 30 Jul 2025 00:00:00 +0700

Indonesia is one of the countries that has a diversity of mango plants. Mango fruit is in great demand throughout the world, but pest control of mango diseases is still not too optimal, one of which is on the leaves. So the automatic recognition of diseases on mango leaves will have a very important role in achieving satisfactory yields. With these problems, this study evaluates models using Convolutional Neural Network (CNN) architecture VGG16 + SVM and VGG alone. The dataset consists of 4000 digital images with seven (7) disease classes and one (1) healthy class. This study shows that the VGG16 + SVM model has a fairly good performance in disease detection on mango leaves with accuracy, precision, recall, and F1-Score values of 99.50%, 99.50%, 99.50%, and 99.50%, respectively.

Implementation of Layered Encryption Using Vigenère and Rail Fence Cipher on Live Chat Systems for Customer Data Security

M. Fathony Ramdhan — Wed, 30 Jul 2025 00:00:00 +0700

The security of digital communication has become increasingly crucial with the growing use of instant messaging services. This study aims to design and implement a live chat system that applies layered encryption using two classical cryptographic algorithms, namely Vigenère Cipher and Rail Fence Cipher, to enhance message confidentiality. Messages sent by users are first encrypted using the ASCII-based Vigenère Cipher with mod 256, and then encrypted again using the Rail Fence Cipher with a zigzag pattern. The testing process considered several parameters such as message length (10–100 characters), variations in key length, number of rails (2–5), and the number of messages sent per minute. The results indicate that the combination of these two algorithms can maintain message security without affecting real-time communication performance. This approach demonstrates that layered encryption using classical algorithms is still relevant for enhancing the security of digital communications in small to medium-scale applications.

The Advanced Feature-Driven Smartphone Price Prediction Using Machine Learning

Numan Shafi, Farzeen Atif, Attroba Atif, Robass Atif — Wed, 30 Jul 2025 00:00:00 +0700

A rapid growth occurred in the smart phone industry owing to the launch of numerous fresh devices across various price categories. It is advantageous to both the device makers and the users to be aware of the relative price of a smartphone which is in relation to its specifications. This research focuses on artificial intelligence for the purpose of making predictions on the price ranges of the devices with some specific attributes such as:- Fingerprint Sensor, storage space, 5G Support, Water Resistance and Number of Sims , Face Lock. The last one is rather significant, as in some regions of the world devices fitted with one or two SIM cards or more are likely to be used most because of the different wide range of connectivity they offer. This methodology includes cleansing, transformation of data and applying suitable techniques from the machine learning. The developed model is precise and accurateand the output information of the model suffices a good number of questions that assist companies on how to set prices and also shape their clients’ buying behaviors.

Website-Based Gold Price Movement Prediction System Using the Long Short-Term Memory (LSTM) Method

Farlin Nurjananti — Wed, 30 Jul 2025 00:00:00 +0700

Gold price prediction is an important aspect in supporting investment decision making amid dynamic market fluctuations. This research aims to find the best hyperparameter combination in building a gold price prediction model using the Long Short-Term Memory (LSTM) method through the Grid Search and Bayesian Optimization tuning approaches. The data used is historical gold price data from Yahoo Finance for the last 10 years (2015-2025) which includes date attributes, opening price, closing price, highest price, lowest price, and trading volume. This study was conducted with two data divisions, namely the 70%:30% and 80:20 ratios, to evaluate the performance of the model to find optimal results. The hyperparameter tuning process includes finding optimal values for epoch, batch size, learning rate, number of neurons, dropout, and optimizer parameters. Model evaluation was conducted using MAE, RMSE, and MAPE metrics and the best results were obtained from tuning using Grid Search at a split ratio of 70%:30% with MAE values of 19.5470, RMSE of 26.5331, and MAPE of 0.93%. The system automatically updates the daily model by using web scraping technique and utilizing Python scheduler. The system was developed based on a website using the Flask framework and has an interactive display consisting of dashboard, historical, and prediction pages. The test results show that the system runs according to its function and the LSTM model is able to predict gold prices with good accuracy. This research shows that proper hyperparameter tuning can significantly improve the performance of the prediction model.

Application of The Extreme Learning Machine to Predict the Amount of Duck Egg Sales (Case Study: Barokah Farm)

Viola Firdayanti, Saiful Bukhori, Yudha Alif Auliya — Wed, 30 Jul 2025 00:00:00 +0700

Barokah Farm is a duck farm that produces two varieties of eggs, salted and fresh duck eggs. In the midst of deal preparation, the challenge often arises in maintaining an sufficient stock of eggs, especially since the production of salted eggs has not been able to fulfil customers' demands. The accessibility of duck eggs impacts transactions, as stock levels must adjust with requests to avoid abundance and deficiencies. This considers the Extraordinary Learning Machine (ELM) strategy to predict the transactions of salted and fresh duck eggs. The purpose of this research is to apply the ELM strategy to forecast the amount of duck egg transactions at Barokah Farm and to survey the error rate or percentage of error. Deals forecasts are conducted employing a testing plot to distinguish ideal hyperparameters. From the test parameters that have been tried, it is found that average value is able to reduce the percentage error to less than 10%, where the least MAPE test was gotten from the number of features 4 of 2.050% for salted eggs and 1.796% for fresh eggs. Using a multiple neurons of 7 with a MAPE value of 0.329% for salted eggs and 0.466% for fresh eggs. While for the data ratio, the best ratio was found to be 80%:20% with a MAPE value of 0.401% for forcasting salted eggs transaction and 0.550% for fresh eggs. It also applied a biner sigmoid activities function with the least MAPE value of 0.032% for salted eggs and 0.524% for fresh eggs.

Implementation of Convolutional Neural Network (CNN) for Watermelon Plant Diseases Using Lenet-5 Architecture

Muhammad Ilham asshidiq, Tio Dharmawan — Wed, 30 Jul 2025 00:00:00 +0700

Watermelon (Citrullus Lanatus) is a horticultural commodity with high economic value that is widely cultivated by farmers in Indonesia, but the productivity of watermelon plants is often hampered by various types of diseases such as anthracnose, leaf spot and viral mosaic which can reduce the quality and quantity of crops. Manual identification of leaf diseases by farmers or agricultural experts is often subjective and takes a short time. Therefore, a plant disease identification method that is fast, accurate and easily accessible is needed. Artificial intelligence technology, especially Convolutional Neural Network (CNN), has proven effective in applying classification and object detection. Lenet-5 architecture is one of the early forms of CNN developed by Yann LeCun which is now widely reused for image classification purposes. The dataset used consists of 4 classes namely Antrachnose, Downy mildew, Healthy leaves and Mosaic virus with a total of 1155 images. The evaluation results of this study obtained the accuracy, recall, precision and F1-score results are 93%, 94% 95% and 94% respectively.

Disease Classification in Cauliflower Plants Using Vgg19 Architecture and Support Vector Machine (SVM) + Lime

Utut Ardiansah, Tio Dharmawan — Wed, 30 Jul 2025 00:00:00 +0700

Cauliflower (Brassica oleracea L. botrytis) is a cool-season vegetable rich in fiber, vitamin B, and phytonutrients that provide significant health benefits including cardiovascular protection and cancer risk reduction. Manual monitoring of plant diseases is extremely difficult and time-consuming, making early disease detection crucial for efficient cauliflower cultivation in the agricultural sector. This study aims to classify diseases in cauliflower plants using VGG19 architecture combined with Support Vector Machine (SVM) and Local Interpretable Model-Agnostic Explanations (LIME). The dataset consists of 7,360 digital images covering three disease types (downy mildew, black rot, bacterial spot rot) and healthy plants. Results show that the VGG19+SVM model achieved 99.86% accuracy, outperforming standalone VGG19 (99.46%). LIME analysis successfully visualized critical areas underlying the model's predictions. These findings demonstrate the effectiveness of combining deep learning and machine learning for plant disease detection, while enhancing model interpretability through visual explanations.