International Journal of Computer (IJC - Global Society of Scientific Research and Researchers, GSSRR)
Not a member yet
459 research outputs found
Sort by
Analyse of Weather Forecasting Models Using LSTM with Noise Removal Methods
Full text linkWeather forecasting is one of the most important fields for all sectors such as transportation (air traffic, marine), industry, agriculture, forestry and even public health sector. The aim of this study is to analyse the weather forecasting models using filtering techniques which can remove the noise involved in the time series weather data. We utilize ten years weather data (2013 to 2022) in Hmawbi region, Yangon, Myanmar. The weather dataset includes maximum and minimum temperatures, humidity, wind speed, cloud amount and weather type. It is taken from the Department of Meteorology and Hydrology, Myanmar. The original data includes noise data. Thus, two noise removal methods, Simple Moving Average and Weighted Moving Average, are used for data cleaning. In this paper, three prediction models are developed by using Long Short-Term Memory (LSTM) with different datasets: one with the original data (without noise removal), one with data cleaned using the Simple Moving Average method, and another with data cleaned using the Weighted Moving Average method. The research goal is to compare and evaluate the performance of these three weather prediction models to determine which one gives the better results. To evaluate these models, Root Mean Square Error (RMSE) and Mean Square Error (MSE) are calculated for each model. LSTM achieves RMSE of 5.398 and MSE of 29.020, SMA achieves RMSE of 3.767 and MSE of 14.841, WMA achieves RMSE of 0.798 and MSE of 0.534. According to the experimental results, it is found that model with data cleaned using the Weighted Moving Average method is lower error rates than other two models and gives the better predicting results than other two models
Cloud Security Architecture for the Automotive Industry: A Framework for Secure Multi-Cloud Deployment
Full text linkThis paper examines the automotive industry since it fundamentally transforms toward Software-Defined Vehicles (SDV) and exponentially increases the reliance on cloud and multi-cloud infrastructures. The attack surface expands in a dramatic way because of a shift to this model. Today cyber risks target centralized cloud backends managing entire fleets not individual vehicles. A comprehensive security architecture should be developed proactively and without delay, as potential threats may escalate, and stringent regulatory frameworks such as UNECE R155 and ISO/SAE 21434 are anticipated to come into effect. In this paper, the Automotive Multi-Cloud Security Framework (AMCSF), which is a conceptual cloud security architecture model for the automotive industry, is proposed based on a systematic literature review as well as comparative analysis of industry practices. ISO/SAE 21434 describes vehicle cybersecurity lifecycle processes the five-layer model would integrate. It works also with modern cloud-oriented technologies like Cloud Security Posture Management (CSPM) and Cloud Workload Protection Platforms (CWPP). The study primarily concludes that risk can be effectively managed in modern automotive ecosystems only if protections are applied synergistically and include both the vehicle itself and its cloud infrastructure. This paper will be helpful to cybersecurity and cloud solution engineers and architects in automotive (OEM, Tier-1/Tier-2), vSOC/DevSecOps teams, cybersecurity and compliance management specialists (CSMS/UNECE R155, ISO/SAE 21434), as well as consultants and executives responsible for designing and operating secure multi-cloud backends
AI-Driven Capacity Forecasting for Last-Mile Logistics
Full text linkThe article examines the features of throughput forecasting based on artificial intelligence in the field of last-mile logistics. The relevance of the topic is driven by the fact that, under conditions of rapid growth in e-commerce and tightening consumer expectations, last-mile logistics faces unprecedented pressure, requiring hyper-accurate forecasting of operational capacity to enable a just-in-time delivery model and meet delivery deadlines without excessive costs. Traditional planning methods demonstrate their inadequacy under conditions of high demand volatility. This work describes an artificial intelligence (AI)-based framework used for forecasting daily resource requirements (personnel, transport) in large-scale logistics networks. The framework is based on a hybrid machine learning architecture that combines regression models for baseline load forecasting and the XGBoost algorithm for detecting and quantifying abnormal demand spikes. A key element of the system is a dynamic capacity buffer algorithm that, in real time, calculates the required reserve of resources to mitigate risks associated with forecast errors. The article analyzes the architecture, implementation methodology, and empirical results, and also discusses the role of such systems as a strategic tool for demand shaping
Maturity Metrics for Engineering Teams in AI-First Application-Development Projects
Full text linkThis study is conducted to define a practical maturity-metrics model for AI-First engineering that links delivery and ML lifecycle to business outcomes, enabling reproducible, responsible, scalable operations across process, platform, team, and product. Comparative analysis of maturity models and delivery indicators (DORA Four Keys) combined with ML-specific metrics (retrain time, drift alerts, explainability coverage). Systematic review of MLOps practices and enterprise cases. SMART-based operationalization, automated collection via unified observability and event telemetry; explicit addition of a product plane to Microsoft’s MLOps model. Triangulation against McKinsey and DORA survey data. The model formalizes maturity across four planes—process, platform, team/culture, and product—mapped to five levels from no MLOps to fully automated operations. It integrates delivery indicators (lead time, deployment frequency, change failure rate, MTTR) with ML metrics (retrain time, drift alerts, explainability coverage) and requires SMART, automated, actionable collection. It further extends the metric set to cover generative AI and LLM-specific behavioural indicators such as factuality, instruction-following, retrieval effectiveness, safety violations and cost per useful token, and introduces repeatable proxy measures for people and culture—including knowledge diffusion, psychological safety and role-ratio tracking—so that human and organisational factors are observable and auditable. Evidence from DORA and McKinsey indicates that teams instrumenting both delivery and ML metrics achieve higher productivity, steadier SLA adherence, and reduced burnout; meanwhile, rapid AI tool uptake coexists with low trust in generated code (only 24% fully trust), underscoring an adoption–resilience gap the model addresses. Clustering metrics by change delivery, ML lifecycle, user/business value, reliability and quality, governance/ethics, and people/culture focuses investment. Operational enablers—event-based telemetry, unified observability, infrastructure as code, and SLO-prioritized alerting—support uninterrupted releases and faster recovery. The framework functions as an audit checklist and portfolio-planning instrument, translating measurements into managed, value-linked actions. Unifies DORA delivery metrics with ML lifecycle indicators across four planes, adding a product layer and operational enablers (event telemetry, IaC, SLO-prioritized alerting) to make metrics actionable for investment
Specifics of Software Quality Assurance in High-Frequency Trading (HFT) Systems
Full text linkThis article addresses how one can combine AI-generated synthetic medical datasets as well as FHIR standard semantic standards and APIs to better enable cross-system interoperability between health care providers. In order to create data sets with the desired level of analytical utility, the author have as a study objective to demonstrate how the generation of synthetics in the context of FHIR resource and transport via the R5 mechanisms of profiles, subscriptions and asynchronous export to NDJSON removes both legal and technical barriers to accessing clinical data. Because data-sharing is complicated by the risk of regulatory noncompliance; a significant amount of clinical data is contained in unstructured documents; and clinical data grows at an exponential rate, the need for this type of solution has increased. This work is innovative in that it brings together three families of generative models: sequential LLM-like models, GANs and diffusion networks. Additionally, the author use a direct mapping pipeline from generative models to FHIR resources and validate profile IG and automated REST tests using built-in systems. To train the generative models on rare nosologies, and to perform load-testing, the author describe an architecture that enables private, secure access to multiple, representative cohorts, which are used to generate the synthetic data. Key Findings: (1) The use of synthetics inside the FHIR shell ensures compatibility with the current infrastructure while reducing legal barriers. (2) They also allow risk-free testing of extreme scenarios and (3) they also reduce sample bias. A rational and practical strategy for accelerating the implementation of analytics in clinics would include a strategy of generating synthetics in a continuous integration/continuous delivery (CI/CD) environment and (2) implementing mandatory validation through profile management. The author believe that the article will provide valuable assistance to medical AI developers, integration solution architects, IT service managers, and regulatory analysts
Strategies for Database Performance Optimization in High-Load Systems: A Review of PostgreSQL and Redis Use Cases
Full text linkThe article presents an expanded analysis of performance-optimization strategies for database systems operating under high load, using PostgreSQL and Redis as representative examples. The study is based on comparing architectural principles of storage systems, query-planning models, mechanisms for handling concurrent operations, and the characteristics of in-memory data processing. It examines differences in DBMS behavior as workload profiles change—from read-dominant scenarios to intensive write patterns and mixed workloads. Particular attention is given to how PostgreSQL’s internal mechanisms, including cardinality estimation, plan selection, and concurrency management, respond to data-volume growth and increasing query complexity. A substantial part of the study focuses on the influence of infrastructural decisions—containerization, automatic scaling, load balancing, distributed queues, and fault-tolerance mechanisms—on the real performance of DBMSs under load. The analysis demonstrates that the stability of data-processing pipelines depends on the coherence between database-level optimizations and cloud-environment parameters, enabling the reduction of latency, improved resilience to failures, and effective operation during peak demand. The practical contribution of the work lies in identifying strategies that allow engineers to design databases and supporting infrastructure as a unified optimization system. The article will be useful for database administrators, developers of high-performance services, distributed-systems architects, cloud-platform engineers, and researchers studying data-processing mechanisms in highly loaded computational environments
Defensive Cybersecurity Preparedness Assessment Model for Universities
Full text linkBroadband and internet access has become readily available to citizens across the globe as a result the recent uptake of fiber connectivity. General Cyber Security threats like malware attacks, social engineering scams and financial frauds have increased. Though numerous security models have been advanced by NIST and ISO standards, but the frightening truth is that escalating cyber-attacks are still on the rise. This is because most existing security analysis tools focus mainly on detecting attacks. Despite the steady flow of security updates and patches, this scenario has led to a continued rise of attack surface in institutions of higher learning where students and staff sensitive information and valuable assets is of high stake. Therefore, the purpose of this study is to establish the factors for effective defensive cyber security in Universities. The study utilized a survey method to collect data from cyber security experts of the sampled universities. The study targeted 27 respondents (ICT experts) from 5 universities both public and private that were purposively sampled in Kenya. 23 questionnaires were returned translating to 85% response rate. This was very sufficient for the study. Correlation analysis was carried and the findings indicated a statistically significant relationship for human factors (87.7%), technology factors (83.5%), and policy factors (83.2%) on defensive cyber security preparedness. Multiple linear regression was also done to predict the extent of the effect of each independent variable on defensive cyber security preparedness. In conclusion, the study noted that, all the three cyber security factors were significant hence there was a need to enhance them so as to improve security against the advancing threat landscape across all sectors especially institutions of higher learning like universities
Application of Linked Lists for Managing Fragmented Memory in the Unity Game Engine: Approaches and Optimizations
Full text linkThis paper examines the problem of managing fragmented managed memory in the Unity game engine. It proposes using linked lists combined with an object pool to minimize garbage?collector pauses and smooth out frame time. The relevance of this work stems from the fact that traditional contiguous?memory containers (arrays, List<T>) under frequent Instantiate/Destroy cycles cause heap fragmentation and abrupt stop?the?world GC passes, leading to noticeable FPS drops even with relatively small allocation volumes. The objective of the study is to systematically evaluate the effectiveness of various linked?list–based pool implementations (the standard LinkedPool<T>, intrusive lists, structural nodes in a NativeContainer) according to the metrics of GC?Alloc, frame time, and heap fragmentation in typical game scenarios involving the spawn of hundreds of prefabs per frame. The novelty lies in a comprehensive comparative analysis of four approaches within a unified Unity testbed, employing detailed profiling (Unity Profiler, Memory Profiler, ProfilerRecorder), and in demonstrating that even the simple use of LinkedPool<T> reduces the number of GC?stops by more than an order of magnitude, while advanced techniques (intrusive lists, NativeList within Burst/Jobs, pool sharding) additionally eliminate hidden allocations and render frame time practically stable. The key findings show that a linked list with object pooling enables O(1) Get/Release operations without massive copying and without requiring a contiguous memory block; the intrusive approach reduces node overhead; and “structural” nodes in a NativeContainer achieve zero GC allocations within Burst and Jobs subsystems. Pool sharding by object characteristics maintains constant complexity under scaling, and validation via profilers confirms zero GC?Alloc per frame and a stable memory?usage curve, making the proposed optimizations practically applicable to mobile and desktop Unity projects. This paper will interest game?engine developers, technical artists, and performance?optimization engineers working on Unity projects
AI Approaches to Software Quality Assessment: From Defect Prediction to Test Coverage Optimization
Full text linkThe article addresses the problem of growing inefficiency in traditional approaches to software quality assurance (QA) under accelerated cycles of continuous integration and delivery (CI/CD). The aim of the study is to present and evaluate a new integrated platform that leverages large language models (LLMs) to automate and optimize key QA processes. The methodology is based on an industrial case study of a system utilizing Claude 4 and Amazon Q CLI for semantic log analysis and predictive test prioritization. The paper presents key quantitative results, including an increase in nightly build stability from ~70% to over 90%, a one-third reduction in regression testing time, and a halving of pull request verification time. Additionally, it is emphasized that the adoption of such solutions enhances testing transparency, improves collaboration between development and QA teams, and accelerates release cycles while ensuring higher product quality. The main contribution of the article is the provision of empirical evidence from a large corporate environment, confirming that modern AI-driven approaches can significantly improve the efficiency, accuracy, and strategic value of software testing. The article will be useful for researchers, quality engineers, CI/CD architects, and DevOps practitioners interested in applying LLMs to optimize testing processes
Unified Benchmark for Evaluating Performance, Bias, and Consistency in LLM Binary Question Answering
Full text linkBinary question answering is central to many real-world applications of large language models (LLMs), such as fact-checking or decision-making support. Yet, despite its prevalence and the high stakes of getting a binary judgment wrong (where an error yields the exact opposite outcome), there are no recent comprehensive benchmarks dedicated to evaluating LLM behavior on this task. To address this gap, we introduce a unified benchmark for assessing binary QA across three dimensions: performance, bias, and consistency. The benchmark is supported by a five-domain dataset augmented with new controlled reformulations of each question, including paraphrases, negations, and answer option variations. Across fifteen state-of-the-art LLMs, we find strong overall performance on the task, with larger and reasoning-optimized models showing better results than the smaller variants. At the same time, we observe pervasive No-leaning bias, universally weak consistency when handling semantically opposite questions, and substantial cross-domain variation. Reading comprehension and multi-hop reasoning topics are handled reliably, whereas numerical reasoning, ethical judgment, and, especially, translation evaluation remain challenging. These findings reveal both the strengths and shortcomings of current LLMs on binary QA, providing researchers with a basis for targeted future improvements while also helping practitioners make informed choices when deploying the models in binary decision contexts