Daily Archives: July 8, 2026

Authors: Egondi Obinga Patrick, Dr. Mwenda Paul, Dr. Ongalo Thomas

Abstract: This study examined the influence of financial transparency on the financial performance of Deposit-Taking Savings and Credit Cooperative Societies (DT-SACCOs) in Kenya, with a particular focus on the moderating role of SACCO size. Drawing on Agency Theory, Stakeholder Theory, and Resource Dependence Theory, the study adopted a mixed-method explanatory design involving 221 respondents drawn from 123 DT-SACCOs regulated by the SACCO Societies Regulatory Authority (SASRA). Quantitative data were analysed using correlation and regression techniques, while qualitative insights were used to contextualise statistical outcomes. Findings revealed that financial transparency has a strong and statistically significant positive effect on financial performance (r = 0.844, p < 0.001), explaining 71.1% of performance variation (R² = 0.711). However, interaction analysis indicated that SACCO size significantly moderates this relationship, with larger SACCOs exhibiting reduced marginal benefits from transparency due to increased structural complexity. The study concludes that financial transparency is a critical governance driver of financial performance, but its effectiveness depends on institutional scale. Policy implications emphasise differentiated governance frameworks based on SACCO size to enhance efficiency, accountability, and sustainability in Kenya’s cooperative financial sector.

DOI: https://doi.org/10.5281/zenodo.21258209

Published by: vikaspatanker

Authors: Maureen Nechesa Murambi, Daniel Khaoya Muyobo, Richard Rono

Abstract: Traditional meteorological models often face challenges in processing large volumes of real-time data and capturing complex nonlinear atmospheric relationships. Recent advances in Machine Learning (ML) have provided powerful tools for analysing weather patterns and improving forecasting accuracy. The paper discusses relevant literature on machine learning algorithms suitable for weather pattern analysis, identifies research gaps and proposes future research directions involving deep learning and hybrid forecasting systems. This paper presents an integrated Internet of Things (IoT) and Machine Learning (ML) model for analysing weather patterns in Bungoma County, Kenya. Historical weather data (2006–2025) from the Nzoia Sugar Factory Weather Station and simulated real-time IoT sensor observations were analysed using Random Forest (RF) and K-Nearest Neighbours (KNN). Data preprocessing included outlier detection using the IQR method, polynomial interpolation for missing values, Min-Max normalization, and feature engineering. The model was trained and evaluated with an Infinite Random Search hyperparameter optimiser (578 configurations, 3-hour window). Performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R²).. The overall average R² across all predicted weather targets was 0.495, with relative humidity at 15:00 achieving R² = 0.836 and maximum temperature achieving R² = 0.629. Comparative evaluation showed that RF consistently outperformed KNN in predictive accuracy, demonstrating the suitability of ensemble learning for nonlinear meteorological datasets. The integration of IoT enabled continuous monitoring and improved decision support for agriculture and disaster preparedness. These findings contribute to the growing body of knowledge on ML applications in meteorology and provide a foundation for developing localized weather forecasting systems in regions with similar climatic conditions.

DOI: https://doi.org/10.5281/zenodo.21257834

Published by: vikaspatanker

Authors: Md Juman Hussan

Abstract: Blood services collect a lot of data. They know who donates, when, how often, and what happens to every unit of blood after that. Most of this data still sits in spreadsheets and databases, doing very little beyond record-keeping. This paper looks at a simple, well-known framework for AI, the same one taught in introductory AI courses, and asks a plain question: what could a blood service actually do with it? Using Australian Red Cross Lifeblood as a case study, this paper walks through four uses of everyday AI: predicting which donors are about to stop donating, forecasting blood stock before shortages happen, supporting donor screening questions, and using generative AI to reach donors in their own language. Real figures from Lifeblood's published donor study, transplant program, and research investment reports are used throughout to ground the discussion in actual numbers rather than hypothetical ones. None of the ideas here need advanced or futuristic technology. They need clean data, a clear question, and a narrow tool built for one job.

DOI: https://doi.org/10.5281/zenodo.21257647

Published by: vikaspatanker

Authors: Tosif Raza Mansoori

Abstract: Customer retention has become one of the most significant challenges faced by modern banking organizations. Due to increasing competition in the financial sector, customers can easily switch from one bank to another if they are dissatisfied with the services provided. Therefore, predicting customer churn has become an important business problem, as retaining existing customers is generally more cost-effective than acquiring new ones. This project presents a Machine Learning-based Bank Customer Churn Prediction Dashboard developed using Python and Streamlit. The objective of the project is to analyze customer information and accurately predict whether a customer is likely to discontinue banking services. Along with prediction, the dashboard provides interactive visualizations and business insights that assist organizations in making informed decisions. The project begins with data collection and preprocessing, where duplicate records and unnecessary attributes are removed. Categorical variables are converted into numerical values using Label Encoding, and numerical features are standardized using StandardScaler. The cleaned dataset is then used to train multiple Machine Learning classification models. Three Machine Learning algorithms were implemented and compared, namely Logistic Regression, Decision Tree Classifier, and Random Forest Classifier. Their performances were evaluated using Accuracy Score, Precision, Recall, F1-Score, and Confusion Matrix. Experimental results showed that the Random Forest classifier achieved the highest prediction accuracy of 86.25%, making it the final model selected for deployment. To improve usability, the trained model was integrated into an interactive Streamlit dashboard. Users can enter customer details and instantly receive churn predictions along with prediction confidence, customer risk level, and business recommendations. The dashboard also includes interactive data visualizations, customer analytics, dataset exploration, feature importance analysis, and model comparison charts. Overall, this project demonstrates how Machine Learning and Data Analytics can support banking organizations in reducing customer churn, improving customer retention strategies, and making data-driven business decisions.

Published by: vikaspatanker