Amid increasingly unpredictable climate shifts and ecosystems undergoing radical transformations due to pollution, various diseases and outbreaks are rapidly adapting.

Among these, skin diseases remain one of the most commonly encountered conditions across diverse demographic groups. Although often perceived as minor, skin diseases can exhibit epidemic characteristics—spreading rapidly and affecting individuals of all ages.

In addition to their epidemic nature, skin diseases are generally difficult to detect. Limited health literacy and relatively inadequate detection methods are key factors contributing to the challenges in managing the majority of skin-related conditions.

Furthermore, the lack of advanced diagnostic tools has compelled many healthcare professionals to rely heavily on clinical experience and subjective visual assessments in diagnosing most skin diseases.

In response to these challenges, Mr. Hadiguna Setiawan, S.E., M.Kom., a lecturer from the Informatics Engineering Study Program at Universitas Dian Nusantara (UNDIRA) conducted collaborative research aimed at developing an AI-based system capable of providing automated early evaluation of skin disease symptoms.

The system was developed using a Deep Learning approach, specifically employing a Convolutional Neural Network (CNN) designed to process and analyze visual data such as images and videos. Fundamentally, CNNs mimic the human visual system by learning to recognize object features and patterns without requiring manual feature extraction.

Despite their widespread application in medical analysis, CNN models are inherently prone to overfitting—a condition in which the model excessively learns fine-grained details from training data, leading to noisy or less generalizable predictions when applied to new datasets. This limitation becomes particularly detrimental when analyzing skin images with complex textures.

To address this issue, the researcher implemented a Dropout mechanism to regulate the data extraction patterns within the CNN architecture. In addition to the deep learning approach, traditional feature extraction techniques such as Local Binary Pattern (LBP) were also utilized. LBP is a robust algorithm capable of capturing detailed image intensity variations and skin texture patterns.

The optimization process began with the collection of 32 batches of sample data, followed by training the CNN model over 50 epochs with Dropout applied to enhance stability and mitigate overfitting.

The results revealed several notable findings. The system was not only capable of accurately mapping skin images but also effectively identifying various types of skin diseases, including Melanoma, Dermatofibroma, Melanocytic Nevi, Vascular Lesions, Basal Cell Carcinoma, Benign Keratosis, and Actinic Keratoses. Moreover, the system provided precise probability distributions for each detected condition.

Performance evaluation demonstrated a validation accuracy of 99.45%, an F1-score of 99.32%, and a validation loss of 0.012. This research represents a significant advancement in both dermatology and artificial intelligence, highlighting that AI models integrating CNN and LBP not only support everyday computational tasks but also enhance health literacy and enable early diagnosis, allowing patients to take preventive measures more effectively.

This study further reinforces the role of Informatics Engineering in developing innovative and sustainable solutions for public health. Universitas Dian Nusantara (UNDIRA) remains committed to cultivating highly competent graduates capable of advancing software engineering and intelligent systems to transform the technology sector across creative industries, public services, and entrepreneurship.

Together with Informatics Engineering at UNDIRA, let us become part of a system that advances the field of information technology toward a brighter and more professional future.