INTERNATIONAL SCIENTIFIC JOURNAL OF ENGINEERING AND TECHNOLOGY (ISJET)

An Approach of AI as a Practice of Escapism

Waralak Vongdoiwang Siricharoen — 2026-05-01

This paper examines the phenomenon of AI-mediated escapism by clarifying its scope, risks, and implications for design. The primary objective is to conceptualize how artificial intelligence reshapes traditional forms of escapism through personalization, immersive interaction, and adaptive feedback. Using a conceptual synthesis approach, this work integrates insights from human-computer interaction, psychology, and media studies. The analysis highlights three key findings: 1) AI intensifies escapism by enhancing interactivity, empathy simulation, and agency; 2) the spectrum of AI-mediated escapism ranges from therapeutic and time-bound uses to problematic patterns associated with misinformation, avoidance, and excessive immersion; and 3) ethical design principles—such as transparency, time-awareness, and emotional boundary safeguards—are essential to mitigate risks. The implications extend to HCI, UX design, and policy, suggesting that responsible frameworks can balance the benefits of AI-driven escapism with its potential harms. Limitations include the conceptual and narrative scope of this review, which does not provide empirical validation. Future research should apply empirical methods to test and refine the proposed framework. Overall, this study provides a structured understanding of AImediated escapism and oﬀers design guidelines for creating safer, more ethical interactive systems.

Artifcial Bee Colony Algorithm with Cosine Similarity on Raspberry Pi Cluster for Enhancing Medical Data Classifcation

Anan Banharnsakun — 2026-05-01

Machine learning plays a very important role in our daily lives. Machine learning is used to solve difficult problems encountered in many fields of study. Especially in medical diagnosis, machine learning is used to support patient diagnosis. In this research, the artificial bee colony approach, a well-known bio-inspired algorithm in the field of machine learning, is proposed to enhance the classification of cancer diagnosis data by combining it with cosine similarity measurements and parallel computing on Raspberry Pi clusters. The performance of the proposed method is validated and compared with other well-known algorithms against the Breast Cancer Wisconsin dataset taken from the UCI machine learning repository. Experimental results indicate that the proposed method achieves superior classification accuracy compared to existing approaches, yielding improvements of up to 6.45%. Furthermore, it can scale eﬀectively to support large-scale medical data for disease diagnosis.

A Hybrid Optimization-Simulation Approach for Supply Chain Network Design under Uncertainty and Highly Perishable Environments

Jirawat Chatchaichalermporn — 2026-05-01

Designing and managing logistics and supply chain networks for perishable products requires careful consideration of product quality degradation over time due to environmental conditions and processing delays. This study proposes a Hybrid Optimization-Simulation (HOS) framework that combines a Mixed-Integer Linear Programming (MILP) model with a Discrete Event Simulation (DES) to design a cost-efficient and quality-preserving distribution network for ﬂoral products. The MILP model incorporates a TimeTemperature Sum (TTS) constraint to account for product perishability, while the simulation model evaluates the network’s operational feasibility under dynamic conditions. A case study of the ﬂoriculture supply chain, involving growers, hubs, and retailers, is used to validate the proposed approach’s performance. The results show that the deterministic MILP achieves the lowest total cost of $163,418, but does not account for uncertainty. In contrast, the proposed HOS outperforms the typical Simulation-Based Optimization (SBO) by 4.78% in total cost and achieves a much shorter computation time. In addition, the HOS framework can produce logistics and supply chain network configurations that balance cost efficiency and quality assurance, providing a robust tool for supply chain design in highly perishable environments.

A Study of Road Tracking Immunity for Autonomous Driving

Chuanxiang Bi — 2026-05-01

With the advancement of deep learning research, autonomous driving technology has become increasingly mature. However, the accompanying issue is that environmental interference may pose safety hazards for autonomous driving, presenting a significant threat. Therefore, this paper aims to evaluate the resilience of autonomous vehicles to environmental interference by studying the fundamental road-tracking task of autonomous driving. Following the model of real autonomous driving vehicles, we constructed a 1:20-scale intelligent model car, Jetracer, and used it to simulate the road-tracking task of autonomous driving. We designed four diﬀerent environments, first collecting varying numbers of image data in the original environment for comparative analysis. Then, we selected ResNet18 and ResNet34 as models for training and loaded them onto Jetracer for testing in the four diﬀerent environments. The experimental results indicate that as the number of images in the dataset increases, the eﬀectiveness of road tracking also gradually improves. Meanwhile, we found that ResNet18 is more suitable as a training model compared to ResNet34. Additionally, Jetracer demonstrates a certain degree of interference resistance, where slight or localized environmental changes do not significantly affect its road tracking performance. However, if there are significant overall environmental changes or new environments are introduced, Jetracer’s road-tracking performance is severely impacted.

A Review on DonkeyCar Autonomous Driving Platform Using Neural Network

Yangyang Li — 2026-05-01

DonkeyCar is a DIY autonomous vehicle platform for exploring AI-driven self-driving technology. Using the DonkeyCar Unity simulator, models can be trained and tested in a risk-free environment before real-world deployment. This study examines the impact of training data and track complexity on model performance. Results show that increasing the number of training images from 1,346 to 5,351 reduced the error rate from 36% to 3%, thereby improving driving stability. Models trained on complex tracks adapted better but required longer training. However, challenges such as sensor noise, environmental uncertainties, and limitations in obstacle avoidance suggest the need for data augmentation, reinforcement learning, and multi-sensor fusion to enhance model robustness. These findings contribute to optimizing low-cost, vision-based autonomous driving systems.

Automated IoT-Based Mushroom Greenhouse Control System for Real-Time Environmental Management via Blynk Application

Chainarong Janthoom — 2026-05-01

This research presents the development of an automated, IoT-based mushroom greenhouse control system for real-time environmental management, implemented
using the Blynk application. The proposed system integrates SHT30 temperature and humidity sensors with an ESP32 microcontroller to continuously monitor environmental conditions and automatically control water misting in mushroom cultivation houses year-round. Real-time data are transmitted to a cloud platform, enabling remote monitoring and both automatic and manual control via a mobile application.

Experimental results show that the system eﬀectively maintains environmental conditions close to the optimal range for mushroom growth, with temperatures between 24.0-29.0 °C and relative humidity levels of 64-76%RH under normal conditions. The system demonstrates reliable adaptive operation by dynamically switching between Automatic Mode and Timer Mode under unfavorable weather conditions, such as rainfall or low temperatures, achieving 100% operational reliability. Sensor performance analysis indicates high measurement consistency, with coefficients of variation below 10% across all test scenarios.

The proposed system reduces manual labor by approximately 70% while improving the precision of humidity control, maintaining it within ±5% RH, and leading to more consistent cultivation performance. This research contributes to sustainable smart farming by providing a low-cost, scalable, and energy-efficient solution for adaptive management of mushroom greenhouses.

Cost Minimization in 3D Bin Packing under Orientation and Operational Constraints: A Case Study in E-Commerce Fulfillment

Anupong Thuengnaitham — 2026-05-01

Inefficient parcel packing in e-commerce fulfillment causes excessive void space, inﬂated dimensional weight charges, and unnecessary dunnage costs. This issue is rightly classified as a three-dimensional bin packing problem (3D-BPP), which is NP-hard and computationally intractable at realistic scales. Current industry practice relies on manual, horizontal-only packing with predefined box sizes, thereby systematically overlooking opportunities to reduce costs through ﬂexible orientation. While extensive literature addresses 3D-BPP, existing research typically applies operational constraints as post hoc checks after geometric optimization, and rarely combines discrete box selection with explicit logistics cost minimization. This study develops and evaluates an orientation-ﬂexible packing framework that integrates fragility protection, weight precedence, and base-support stability into the heuristic placement process, rather than filtering out infeasible solutions after packing. The framework selects from a set of discrete, predefined box sizes and minimizes the total logistics cost, including packaging and dimensional weight charges. Three strategies—a horizontal-only baseline, First Fit Decreasing (FFD), and Best Fit (BF) with orientation flexibility, are systematically compared using simulation experiments on 600 real-world e-commerce orders from a personal care products fulfillment operation. Results demonstrate that BF achieves a 4.47% reduction in total logistics cost, 9.31% reduction in void space, and 9.31% reduction in dunnage cost relative to the baseline. At the same time, FFD is approximately eight times faster than BF, with moderate efficiency gains. Cross-validation confirms solution stability across heterogeneous subsets of orders, and sensitivity analyses verify robustness to variations in support thresholds, dimensional-weight pricing, and box-material costs. These findings provide actionable guidance for context-dependent strategy selection, demonstrating that integrating operational constraints into packing heuristics do not degrade cost performance and remain computationally feasible for real-world e-commerce fulfillment.

Effect of Intermittent Photoperiod on Yield and Light-Use Efficiency of Common Ice Plant (Mesembryanthemum crystallinum L.) under Controlled-Environment Conditions

Kornlawat Tantivit — 2026-05-01

This study evaluated the eﬀects of diﬀerent photoperiod regimes: Long-day (16 h light/8 h dark), short-day (8 h light/16 h dark), and intermittent photoperiods, including three intermittent regimes (8/8 h, 4/4 h, and 10 min/10 min) on growth performance and biomass productivity of Mesembryanthemum crystallinum L. under controlled greenhouse conditions. Plants grown under intermittent photoperiod regimes exhibited significantly greater fresh weight, dry weight, and plant height than those grown under continuous short-day or long-day conditions (p < 0.05). These results indicate that temporal redistribution of light exposure can enhance growth efficiency,
potentially by improving the coordination between photosynthetic activity and endogenous circadian regulation. Moreover, intermittent photoperiod suggests potential to improve energy-use efficiency without compromising biomass production. The findings demonstrate that intermittent photoperiod management is a promising strategy for the sustainable cultivation of M. crystallinum in smart greenhouses and controlled-environment agriculture systems.

Face Photo and Sketch Synthesis Using CycleGAN: The Role of Residual Blocks and Spectral Normalization

Nuntipat Phisutthangkoon — 2026-05-01

This study investigates the problem of face photo and sketch synthesis using CycleGAN, with a particular focus on architectural choices that affect model performance. A major challenge in this field is the limited availability of paired face photo-sketch datasets, which makes it difficult to develop and evaluate eﬀective models. To address this limitation, this research demonstrates how to generate face sketches and photos from small datasets, oﬀering a practical solution for expanding existing data collections in future applications, particularly in law enforcement and forensic sketch analysis. Four CycleGAN configurations were designed by varying the number of residual blocks in the generator and applying spectral normalization to the discriminator. Experiments were conducted on the CUHK Face Sketch Database, with quantitative evaluations using SSIM and PSNR, as well as qualitative assessments via human visual inspection. The results show that increasing generator depth and incorporating spectral normalization in the discriminator both improve image quality and training stability, although these benefits come at the cost of increased training time. Notably, the configuration with ResNet9 and spectral normalization consistently achieved the highest scores, with SSIM and PSNR values of 0.702 and 18.53 for sketch-to-photo translation, and 0.702 and 17.35 for photo-to-sketch translation, and produced the most visually convincing results in both tasks. These findings highlight the importance of architectural design choices for eﬀective face photo-sketch translation and provide guidance for future work in related applications.

A Hybrid of Syntactic Structure and Contextual Semantics for Enhanced Product Review Sentiment Analysis

Damrong Sattayawaksakul — 2026-05-01

With the rapid growth of e-commerce platforms, analyzing customer reviews has become essential for understanding product perception and supporting data-driven decision-making. Traditional sentiment analysis methods often rely solely on sequential text representations, which may overlook the syntactic structures and contextual nuances present in customer feedback. This study aims to develop an interpretable hybrid framework that eﬀectively captures both semantic meaning and syntactic dependencies to enhance sentiment prediction accuracy. The proposed model combines contextual embeddings from SentBERT with syntactic features extracted via Graph Feature Fusion. The process involves preprocessing customer reviews, extracting aspect-based semantic representations using SentBERT, and constructing syntactic graphs via dependency parsing. These features are merged in a fusion layer and refined using supervised contrastive learning to improve class separability in the sentiment space. SHAP Explainable AI is integrated to provide human-interpretable explanations for the sentiment predictions. The hybrid model achieves a sentiment classification accuracy of 97%, outperforming baseline methods, including BERT (93%) and GCN (88%), as well as classical machine learning algorithms. These findings highlight the effectiveness of integrating syntactic and contextual features in sentiment analysis. The framework can be applied to real-world e-commerce platforms to enhance automated review analysis, improve customer service insights, and support product development strategies.