All of the articles in this proceedings volume have been presented at the 2026 11th International Conference on Sustainable Information Engineering and Technology during September 7 – September 8, 2026 in Malang, Indonesia. These articles have been peer reviewed by the members of the Reviewer and approved...

Text-to-image (T2I) generative diffusion models have made remark- able progress in single-concept personalization. However, extending customization to multiple concepts simultaneously introduces critical challenges, including identity bleeding (where visual features of distinct subjects merge) and attribute...

Counterfactual explanations help users understand image classifiers by showing how an input would need to change to alter a model prediction. Pixel-space adversarial methods achieve high flip rates, but their perturbations are often imperceptible and semantically meaningless. Latent-space generative...

Abstract. Image restoration pipelines often depend on degradation-dependent parameters, such as denoising strength and data-consistency weights, that are fixed before inference or selected through computationally expensive grid search. This paper proposes a lightweight sequential parameter scheduler...

This paper presents a controlled empirical benchmark of data selection for low-shot image classification under a limited annotation budget. Given a small labelled training set and an additional unlabelled candidate pool, we investigate which samples should be selected to improve downstream classification...

Deep learning models heavily rely on large-scale datasets to achieve robust generalization. In low-resource scenarios, models are highly susceptible to overfitting, necessitating strong regularization techniques such as data augmentation. While automated augmentation methods like AutoAugment and RandAugment...

CPU deployment requires inference frameworks that balance latency, throughput, memory, and accuracy. This paper presents a controlled comparison of PyTorch, ONNX Runtime, and NCNN on two representative workloads: real- time object detection with YOLOv12n and monocular depth estimation with DepthAnythingV2...

Driver drowsiness is a major contributor to road traffic accidents, highlighting the need for accurate and computationally efficient detection systems suitable for real-time deployment on resource- constrained edge devices. This study presents a comprehensive evaluation of YOLO11n optimization strategies...

Cardiovascular signals such as electrocardiogram and photoplethysmography are increasingly analyzed using deep neural networks after being transformed into time-frequency images. Although convolutional models can achieve strong classification performance, their predictions remain difficult to interpret,...

Text-based mental health counseling is increasingly adopted due to its accessibility and anonymity; however, the absence of nonverbal cues makes emotion recognition challenging, particularly under high conversation volume and class imbalance. This study evaluates BERT, BiLSTM with GloVe embeddings, and...

The rapidly expanding body of biomedical information poses a challenge for healthcare professionals in accessing accurate answers, particularly given the limited availability of Indonesian-language literature. The use of Large Language Models (LLMs) for biomedical question answering systems also faces...

Natural Language Generation (NLG) enables generative models such as GPT-2 for Indonesian poetry generation, yet challenges re- main in fluency and diversity for low-resource languages. This study analyzes the effects of prompt variations (zero-shot, one-shot, and few-shot), hyperparameters, and fine-tuning...

Host-based intrusion detection uses host activity, including system-call sequences, to distinguish intrusive behavior from normal execution. Although pretrained Bidirectional Encoder Representations from Transformers (BERT) models offer several architectural and computational profiles, their relative...

Higher education institutions (HEIs) increasingly depend on information systems to manage growing volumes of academic, administrative, and accreditation-related data, yet the sustainability dimension of this data management (efficiency, governance, and environmental responsibility) remains underexamined,...

Emotions play a critical part in everyday life since it affects decision-making, behavior, and human interactions with systems. As a result, emotion recognition has become an important topic in affective computing. Among various physiological signals, electroencephalography (EEG) is widely used for emotion...

Multimodal emotion recognition benefits affective computing applications such as companion robots, driver monitoring, and assistive healthcare, yet most audio-visual systems rely on heavy backbones and static fusion mechanisms that are unsuitable for edge devices or cannot adapt to input-dependent modality...

Excessive use of personal digital devices among university students raises concerns regarding environmentally responsible digital consumption. Built-in digital detox features provide system-level tools that support post-adoption digital behavior regulation in higher education contexts. This study examines...

Stress classification using electroencephalogram (EEG) signals has become an important research direction because EEG can represent changes in brain activity related to psychophysiological conditions. However, EEG signals are non-stationary, and analysis using long observation segments can reduce the...

Disaster response coordination in smart cities remains fragmented across agencies, shown by opaque resource allocation and responsibility deficits. This paper proposes a Geospatial Blockchain Disaster Response System (GeoBlock-DRS), combining smart contracts with Geographic Information System (GIS) spatial...

Digital forensic investigations generate tens of thousands of artifacts originating from heterogeneous sources. This makes incident timeline reconstruction both time-consuming and error-prone. This study develops and evaluates an automated five-step pipeline for ontology- based digital incident timeline...

The growth of internet access in Indonesia, reaching 72.78% in 2024 according to the Central Bureau of Statistics (BPS), has become one of the key factors behind the increasing spread of hoax news. The Ministry of Communication and Digital (Komdigi) recorded 12,547 hoax content cases from August 2018...

Large Language Models (LLMs) have demonstrated remarkable capabilities across various natural language processing tasks; however, their performance is highly influenced by the phrasing of the provided prompts. This study evaluates the robustness of the Flan-T5 model against prompt variations on three...

Security monitoring fragmentation in heterogeneous service infrastructures running SMB, FTP, and HTTP simultaneously creates a wide attack surface. Per-service log isolation prevents timely identifica- tion of multi-vector attack patterns. This paper presents a documented, reproducible reference architecture...

The digital game industry has experienced rapid growth driven by increasing demand for digital entertainment, giving rise to online distribution platforms such as Steam that provide review features enabling players to share their gaming experiences. Player reviews accumulated in large volumes represent...

This paper evaluates on-device abstractive text summarization with a small Large Language Model (LLM) on consumer hardware, focusing on the impact of weight quantization on summary quality and computational efficiency. We conduct a paired comparison of Qwen2.5-0.5B-Instruct under three precision configurations...

Cardiac arrhythmias are a leading cause of global mortality, and ECG is the primary non-invasive detection tool, yet manual interpretation is slow and prone to inter-observer variability. Deep learning models trained on large-scale ECG datasets face unstable convergence and degraded generalization due...

Most prior electrocardiogram (ECG) deep-learning studies focus on classification performance and are evaluated on server-class GPUs, leaving the accuracy-efficiency trade-off on edge devices under- characterized. This study compares four 1D Convolutional Neural Net- work (1D-CNN) variants—baseline 1D-CNN,...

Shrimp aquaculture is a major contributor to global seafood production, where accurate monitoring of shrimp size and weight is essential for feeding management, growth evaluation, and harvest planning. In modern farming conditions, shrimp size is commonly assessed through manual sampling, which requires...

This research examines the gaps in programming education literature by analysing the discrepancies between the frequency of programming practice, levels of grit, and the vocational learning outcomes of computing students. Adopting a quantitative correlational approach, the research assessed 87 students...

Bitcoin price direction classification is challenging due to high volatility and complex temporal dynamics. This study investigates the contribution of four technical indicator categories-Momentum (MOM30), Relative Strength Index (RSI14, RSI30), Stochastic Oscillator (%K30, %D30, %K200, %D200), and Trend-Following...

Risk-based audit prioritisation requires a model that ranks officials before an investigation begins, not one that recognises officials already charged. This paper reformulates the prediction of local-official corruption in Indonesia as a strictly prospective task and evaluates it under leakage controls...

Indonesia has set a Nationally Determined Contribution (NDC) target to reduce greenhouse gas emissions by 32% or approximately 834 million tons of CO₂e by 2030. To achieve this, the government has implemented a carbon tax policy using a cap-and-tax approach to limit emissions and encourage a transition...