International Journal of Emerging Trends in Science and Technology

Fraud Detection Combating Mobile Money Fraud in SMS Messages Using Machine Learning

Panji Msowoya — 2024-07-29

This paper presents a machine learning-based fraud detection model to combat the SMS based mobile money fraud. The systems use an XGBoost classifier to process SMS message content real-time and boast a high level of accuracy with predicting malicious intent. Built using Flask framework, the system allows administrators to monitor in real-time all reported alarms for specific transactions regarding fraud and keep statistical records. Future research includes improving the interpretability of models and making them both more scalable as well as robust to adversarial attacks for better fraud detection in mobile money services.

Adaptive Power System Load Frequency Control based on Metaphorless Optimizers

Biobele Alexander Wokoma — 2024-10-29

This research proposes an alternative Load Flow Controller (LFC) to power systems stability studies based on the population-oriented mathematical optimizer called the PI-Rao. This approach is applied to stabilizing frequency deviations in single-area power systems and compared with a basic PI controller. The research employed a dynamic systems model for a singular area system to capture the dynamics of an LFC. The system is composed of two parts: Part 1 is the LFC optimization systems part which makes a global search for the optimal P & I factors based on some prespecified frequency fluctuation. In contrast, Part 2 is the METARPHOLESS-PI part which uses a population-oriented mathematical logic. The presented optimization in (Part 1) follows classical methods inspired by evolutionary approaches to search for the optimal set of fitting parameters. The minimization of area control error (ACE) is considered an objective function involving the minimization of an Integral Time multiplied Absolute Error (ITAE). The results of simulations have shown the superiority of the proposed solution considering the loading changes of 0.02p.u to 0.1p.u and at intervals of 0.02p.u. Thus, the PI-Rao should serve as another potential solution for power systems LFC applications. The current challenges and future research directions for the existing and prospective projects of the power system operators seek microprocessor-based fault analysis solutions.

P ¹ NP: A Formal Proof

Ali Mahdoum — 2024-05-07

According to the conjecture that P ¹ NP, we recall in this paper that class NP includes P, NP-intermediate and NP-complete problems (some of Co-NP and NP-hard problems also belong to NP). It is obvious that if a single problem belonging to NP is formally proved non-polynomial, then P ¹ NP no longer remains a conjecture but rather becomes a formal statement. In this purpose, we formally prove that the Graph-isomorphism problem (belonging to class NP) is non-polynomial time, which leads that P ¹ NP is a formal statement, not a conjecture.

A Pandemic Predictive Model with Convolutional Neural Networks and Deep Reinforcement Learning using Simulated Partial Differential Equations Data.

Sai Nethra Betgeri — 2024-02-29

Detecting the spread of pandemics will greatly reduce human and economic loss. Existing Epidemiological models used for coronavirus disease 2019 (COVID-19) prediction models are too slow and fail to capture epidemic development thoroughly. This research presents a Physics-based Machine Learning Architecture (PMLA) to improve the processing speed and accuracy of epidemic forecasting governed by susceptible–exposed–infected–recovered–deceased (SEIRD) model equations. The dynamics of the epidemic were extracted using Convolutional Neural Networks (CNN) and Deep Reinforcement Learning (Deep RL) from data simulated with Partial Differential Equations (PDEs). The PMLA accuracy is measured using mean squared error. The PMLA prediction model enhances the ability of health authorities to predict the spread of COVID-19 in real time efficiently and effectively.

Keywords: COVID-19, Convolutional Neural Networks, Deep Reinforcement Learning, Partial Differential Equations, Machine learning, Finite Element Method.

Forecasting Population demographics in Lilongwe city: Leveraging Prophet and Time series analysis Techniques

Destiny Mwafulirwa — 2024-02-04

Population demographics provides grounds for forecasting due to complex nature of various editions of population figures. The accurate prediction of population demographics is pivotal for urban planning, resource allocation, and the development of effective policies. In this paper, we utilize historical demographic data collected over time to forecast future demographic trends, enabling informed decision-making by local authorities and urban planners. The study begins by gathering and preprocessing a comprehensive dataset that encompasses various demographic variables, such as age, gender, income, education level, and occupation, among others. We employ Prophet, a robust forecasting tool, with diverse time series analysis methods. It examines historical demographic data, refining models using Prophet’s flexibility and traditional time series techniques like ARIMA and exponential smoothing. By leveraging Prophet and time series analysis, this paper aims to offer accurate forecasts of population dynamics, age structures and migration trends, providing valuable insights for urban planning and policy formulation.

Keywords: population demographics, ARIMA, prophet, time series analysis.

Empirical Study of the Relationship between Capital Formation and Saving in Rwanda: An Econometric Approach

Ngabo Yisonga Matabaro Roch — 2024-01-04

The objective of this study was to investigate the empirical relationship between capital formation and saving in the case of Rwanda. The empirical literature review reveals that saving has a positive effect on capital formation. Results of the research obtained using the Ordinary Least Square method show that in Rwanda there is positive correlation between capital formation and saving. Saving is used to finance investments useful for the development of the country.

In the equation formulated, endogenous variable was capital formation while exogenous variables were saving, foreign direct investments and money supply. It has been demonstrated that coefficient of saving is positive. In Rwanda, an increase 1% of saving ends at an increase of gross capital formation of 1.271241%. Analyses done end at results that estimated coefficients are statistically significant at 5% margin of error i.e. 95% level of confidence. The relative probability for LSAV is less than 0.5. R-squared is 0.980003. This goes to mean that exogenous variables selected explain Gross Capital Formation at 98% and the model if fitted. Therefore, the government of Rwanda should make supplementary effort to encourage saving in the country so to boost the economy through more investment in order to meet its major objective, i.e the economic and social development.