Computer Science / Parallel Computing
Parallel Merge Sort with Load Balancing
This technical research article investigates the performance limitations of conventional parallel merge sort and proposes a load-balanced alternative designed to improve processor utilisation in distributed-memory parallel computing systems. Traditional parallel merge sort progressively reduces the number of active processors during successive merge stages, causing many processors to remain idle and reducing the performance benefits of parallelisation. The proposed approach addresses this limitation by ensuring that all processors continue participating throughout the merging process. Parallel_Merge_Sort_with_Load_B… The paper first explains the conventional parallel merge-sort process, in which data are locally sorted before processors are paired for a series of merging operations. At every subsequent stage, the number of participating processors is halved until only one processor remains responsible for the final merged list. This results in poor processor utilisation and increasing workload concentration. Parallel_Merge_Sort_with_Load_B… The proposed load-balanced parallel merge sort distributes each partially sorted list across multiple processors so that every processor maintains approximately the same number of keys throughout execution. Processor groups use histograms and boundary values to determine how data should be redistributed during merging. Histogram-based partitioning reduces unnecessary data movement, while an index-swapping mechanism is introduced to avoid transferring large blocks of keys when logical processor reassignment can achieve the same result more efficiently. Parallel_Merge_Sort_with_Load_B… Parallel_Merge_Sort_with_Load_B… The algorithm was implemented in C using MPI and experimentally evaluated on a Cray T3E parallel computer and an eight-node PC cluster. Testing considered both uniform and Gaussian key distributions. Results show that performance improvements increase as processor count grows, although communication and histogram-management overhead can reduce benefits for small workloads. Parallel_Merge_Sort_with_Load_B… The proposed technique achieved a maximum merge-phase speedup of 9.6 on a 32-processor Cray T3E and 2.3 on an eight-node PC cluster when processing four million keys. The study concludes that distributing approximately equal workloads across processors can substantially improve parallel merge performance and may also be applicable to related parallel sorting algorithms. Parallel_Merge_Sort_with_Load_B… Overview word count: approximately 330 words. For your portal, I would not label this as university coursework unless you also have the actual assessment brief that uses this paper. This PDF itself only establishes a published academic paper and the authors’ Korea University affiliation.
Read Model Answer →
Machine Learning / Cloud Computing / Artificial Intelligence
4,004 words
Cloud-Based Machine Learning for Financial Fraud Detection
This Machine Learning on Cloud assessment requires students to design, implement and critically evaluate a cloud-oriented machine-learning solution for financial fraud detection. Working as a group, students address a scenario in which a financial-services organisation requires an automated model capable of detecting fraudulent transactions, reducing financial losses and improving customer security. The project combines machine-learning development with critical evaluation of cloud infrastructure, data preparation, model performance and responsible AI considerations. NUL - LD7187 -Assessment Brief … The project begins with a cloud feasibility study comparing at least two major machine-learning platforms such as Microsoft Azure, Amazon Web Services and Google Cloud Platform. Students evaluate factors including performance, scalability, cost, compliance, system integration and vendor lock-in before providing a justified recommendation. The project then moves into Exploratory Data Analysis, where patterns, anomalies and correlations within the fraud dataset are investigated using visualisations such as heatmaps, histograms and boxplots. NUL - LD7187 -Assessment Brief … A substantial part of the assessment focuses on data preprocessing and class imbalance. Students are expected to clean and transform the dataset, apply scaling and encoding, perform feature engineering and investigate techniques such as SMOTE, undersampling and cost-sensitive learning. These decisions must be justified in terms of their potential effect on predictive performance. NUL - LD7187 -Assessment Brief … Students must then select and train at least two machine-learning models. Suggested algorithms include Logistic Regression, Random Forest, XGBoost and Neural Networks. Appropriate cross-validation and hyperparameter-tuning procedures should be applied, followed by systematic evaluation using precision, recall, F1-score, AUC and precision-recall curves. Supporting visualisations should include confusion matrices, ROC curves and feature-importance analysis. NUL - LD7187 -Assessment Brief … The final component addresses professionalism and ethics in cloud-based AI, including bias, fairness, transparency, data privacy and environmental sustainability. Overall, the project integrates cloud-platform selection, exploratory analytics, preprocessing, imbalanced-data handling, predictive modelling, model evaluation and ethical AI into an applied financial fraud-detection solution. NUL - LD7187 -Assessment Brief … Note: the uploaded brief does not explicitly name a referencing system. If your portal requires a selection, I would use Not specified rather than assume Harvard.
Read Model Answer →
Machine Learning / Cloud Computing
3,984 words
Machine Learning on Cloud: Fraud Detection Model Design, Training and Evaluation
This postgraduate group project focuses on the design, development and critical evaluation of a cloud-based machine learning solution for financial fraud detection. The scenario involves a financial services organisation seeking to detect fraudulent transactions in order to reduce financial losses and improve customer security. Students are required to analyse an appropriate dataset, develop a machine learning solution, evaluate its effectiveness and critically consider the suitability of cloud technologies for deployment. The project begins with a cloud feasibility study, requiring critical comparison of at least two major machine learning platforms such as Microsoft Azure, Amazon Web Services and Google Cloud Platform. Evaluation criteria include performance, scalability, cost, compliance, integration and vendor lock-in, followed by a justified recommendation. Students then conduct exploratory data analysis to identify patterns, anomalies and correlations using appropriate visualisations such as heatmaps, histograms and boxplots. A substantial component addresses data preprocessing and class imbalance. Students are expected to clean and transform the data, apply scaling and encoding, perform feature engineering and investigate approaches such as SMOTE, undersampling and cost-sensitive learning. Each preprocessing choice must be justified in terms of its potential impact on model performance. Students must select and train at least two machine learning models, with suggested approaches including Logistic Regression, Random Forest, XGBoost and Neural Networks. Model development incorporates cross-validation and hyperparameter tuning. Evaluation uses fraud-relevant measures including Precision, Recall, F1 score, AUC and precision-recall curves, supported by confusion matrices, ROC curves and feature-importance visualisations. The project concludes with critical consideration of professional and ethical issues in cloud-based AI, including bias, fairness, transparency, data privacy and sustainability. The overall assessment therefore integrates cloud-platform evaluation, machine learning development, imbalanced classification, model evaluation and responsible AI practice. Overview word count: approximately 340 words.
Read Model Answer →