International Journal of Innovations in Science & Technology
Not a member yet
813 research outputs found
Sort by
CLFT: An Optimized Hybrid Cross-Layer Fusion Transformer for Accurate Fake Profile Detection on Social Media
Get PDFThe rapid increase of fake profiles on social media platforms has raised significant concerns regarding online authenticity, user trust, and digital security. Despite various efforts to combat this issue, existing detection methods often fall short due to the evolving nature of fake profiles and the noisy, high-dimensional data involved. This study proposes an optimized Hybrid Cross-Layer Fusion Transformer (CLFT) for detecting fake profiles by analyzing behavioral metadata. The CLFT architecture integrates multi-stage attention mechanisms, including Cross-Layer Fusion Attention (CLFA), Sparse–Dense Hybrid Attention (SDHA), and Temporal-Behavior Embedding Blocks (TBEB), to effectively capture both short- and long-term dependencies in user activities. The model hyperparameters were optimized using the Bayesian Optimization Hyperband (BOHB) framework. Experimental results on a real-world social media dataset show that the proposed model outperforms traditional machine learning techniques and previous Transformer-based models, achieving an accuracy of 99.10%, precision of 99.89%, recall of 99.55%, and an F1-score of 99.72%. Furthermore, the attention mechanisms enhance interpretability by emphasizing the most influential behavioral features, contributing to the model’s transparency and reliability. The findings highlight that Transformer-based models, especially the CLFT, provide a scalable and efficient solution for fake profile detection in noisy environments, with important implications for enhancing social media security. The study emphasizes the need for interpretability in automated detection systems, fostering trust and ensuring better user engagement and platform integrity
Impact of Different Feature Engineering Techniques for Better Classification of Diverse Crops with Sentinel-2 Imagery
Get PDFObserving a large area of Earth\u27s surface using remote sensing has made our work very easy in order to monitor changes. This revolutionary tech can help us make big decisions on time. For this purpose, Sentinel-2 imagery is considered to be perfect since the imagery provided by this satellite is easily available https://scihub.copernicus.eu/ website. The European Space Agency (ESA) and the European Union (EU) have created the Copernicus Program, which includes the Sentinel-2 satellites that use onboard multispectral scanners to effectively monitor the Earth’s surface. This program has contributed significantly to the production of Sentinel-2 multispectral products, which provide high-resolution satellite data for monitoring land cover and use. The Sentinel-2 constellation is the second set of satellites in the ESA Sentinel missions, with the primary goal of land cover/use monitoring. Besides the availability of imagery, Sentinel-2 temporal resolution is 5 days, which helps in quick observation. In this manuscript, we have used different feature engineering techniques on our dataset in order to observe their performance and importance for better classification of diverse crops. We have achieved an overall accuracy of 99% after extracting important information from the dataset and applying a random forest and a gradient boosting classifier. The data set used for this research work was collected by surveying diverse crops in the region of Harichand, which is located North-South of Charsada District in Khyber-Pakhtunkhwa, Pakistan. The detailed Explanation of our Work and proposed methods is discussed in this article
A Framework for Fraud Detection in Banking Transactions Using Machine Learning and Federated Learning
Get PDFThe digital banking revolution has transformed financial services to make payment faster, more convenient, and borderless. But with this revolution came an abrupt increase in fraudulent transactions through credit cards that threatening both the financial institutions and the customers. While conventional fraud detection mechanisms are not capable of addressing new-generation fraud patterns, there is an increasing demand for intelligent, adaptive, and secure solutions with high precision without any data privacy compromise. Proposed model leverages four machine learning models, Linear Regression, Decision Tree, Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN). LSTM and CNN are used due to their power in learning complicated sequential and feature-based patterns, with Decision Tree and Linear Regression added due to their ease, quick execution, and interpretability. Every model is locally trained on partitioned banking datasets for each simulated client. Model parameters are combined with the Federated Averaging (FedAvg) algorithm to create a globally shared fraud detection system. Experimental testing was conducted on a real-world banking transaction data set published in a non-IID manner to mimic real-world client situations. The federated learning paradigm achieved encouraging results: CNN and LSTM models achieved detection accuracy rates of over 95%, with outstanding performance in the detection of hidden or time-series-based fraud patterns. The Decision Tree model also achieved steady performance at 91% accuracy, and Linear Regression achieved a reasonable baseline at 88%. These results indicate that even simple models, when used in a collaborative federated environment, can contribute meaningfully to fraud detection. This research contributes to the body of research supporting federated banking solutions and fills a significant gap by demonstrating how several ML models can coexist and collaborate in a decentralized setup for fraud detection through credit card transactions
IoT-Driven Gas Safety: Combining Dual-Sensor Technology and Cloud Integration for Automated Risk Mitigation
Get PDFA gas leak in a home can be very dangerous and cause accidents or illness if it is not found soon enough. Many existing gas detection systems cannot avoid false alarms and delays, which means better, real-time systems are needed. A system that uses an ESP32 microcontroller, two sensors (MQ6 for high sensitivity and NDIR for confirmation), and detects gas leaks using the Internet of Things (IoT) is presented in this paper. The methodology of the system includes simulating sensor readings, code within the microcontroller, and MQTT cloud messages at gas concentrations running from 0 to 10,500 ppm. The simulation adds both sensor noise and delays from the network to reflect real life, as alarms are sounded only after both sensors agree. Tests showed the system stays true to zero false alarms and has detection rates above 95% up to 100% when gases reach over 5500ppm. Furthermore, MQTT provides consistently low communication latency of 26 to 32 milliseconds, which helps make responding to emergencies nearly real-time. The research introduces a new IoT approach that manages accuracy, dependability, and speed for residential gas safety, validated through detailed simulation experiments
A Signal-Decomposed Ensemble Forecasting and Classification Framework for Household Power Consumption: An STL-Inspired Machine Learning Approach
Get PDFAccurate short-term forecasting of residential power consumption is crucial for smart grid stability, real-time energy optimization, and personalized demand-side management. Traditional time-series and standalone AI models often struggle with the nonlinear, nonstationary, and noise-sensitive nature of high-resolution household load data. Unlike existing models, this study introduces an STL-based residual decomposition fused with lag-aware ML forecasting and threshold-based classification under real-world conditions. To address these challenges, this study proposes a novel STL-inspired decomposition framework integrated with four machine learning models, i.e., Least Squares Boosting (LSBoost), Bagging, Support Vector Regression (SVR), and Multilayer Perceptron (MLP), for forecasting and classification of normalized household energy consumption. The methodology begins with robust preprocessing, including IQR-based outlier removal and min-max normalization, followed by STL-like decomposition into trend, seasonal, and residual components. Lag-based features from the residual signal are used for forecasting via the selected ML regressors. Final predictions are reconstructed and threshold-classified into OK/NOT OK categories to simulate alert-based power decision scenarios. Experimental validation on the UCI Household Power Consumption dataset reveals that SVR achieves the best trade-off among all models, with RMSE = 0.0267, MAE = 0.0193, MAPE = 12.5%, and Pearson correlation coefficient = 0.846. For classification performance, SVR also attains an AUC of 0.941 and a binary classification accuracy of 93.7%. The synergy between STL decomposition and residual-based modeling not only improves regression accuracy but also facilitates threshold-aware classification with high interpretability. Additional visual diagnostics including forecast overlays, residual histograms, ROC curves, and Q–Q plots demonstrate the model’s interpretability and robustness. The proposed ensemble framework not only enhances prediction accuracy but also ensures practical deployment feasibility through threshold-aware decision modeling
Design and Implementation of a Multi-Strategy Algorithmic Trading Bot
Get PDFThe financial markets require speed and accuracy, and thus, the quick take-up of algorithmic trading systems has ensued. This study presents a hybrid trading bot based on machine learning algorithms and technical indicators such as Moving Average (MA) and Relative Strength Index (RSI). The integration of Random Forest significantly improved signal accuracy and reduced false positives. Back testing over 1 year showed a win rate of 73.2% and a return on investment (ROI) of 42.5%, confirming the effectiveness of the hybrid model. The bot is designed to analyze the market in real-time, and it makes trades autonomously, regulates risk, and adjusts to volatile markets
Predictive Modeling of Hospital Waste Generation Using Machine Learning Based on Patient Inflow
Get PDFEffective hospital waste management is a key to the security of the environment and provision of healthcare hygiene. This paper develops a predictive analytics model to forecast the amount of daily hospital waste generated based on patient inflow using a linear regression model. Real-time data from 60 days were gathered in a tertiary healthcare institution, which informed the number of patients and the resultant waste in kilograms. The model obtained the R value of 0.88 in the training and 0.81 in the validation datasets and a Root Mean Square Error (RMSE) of 130.52 kg. The predictor of patient volume was established as significant through statistical validation via ANOVA, and the model was found to be within the key regression assumptions through the residual analysis. The findings emphasize that predictive modeling within a hospital waste planning system is viable, and a cost-efficient and explainable option can be used in operational forecasting. The offered method contributes to enhanced resource distribution, risk aversion, and adherence to the sustainable healthcare objectives
Impact of Peri-Urban Agriculture on Food Self-Sufficiency of Faisalabad: Exploring the Contribution of Per-Urban Agriculture to Sustainable Food Systems in Faisalabad
No full textIntroduction/ Importance of Study: Peri-urban agriculture plays a vital role in enhancing food self-sufficiency and improving nutritional outcomes, particularly in growing cities like Faisalabad, Punjab. This study assesses how it affects local production, household food supply, and stakeholder integration along the urban–rural interface.Materials and Methods: Data were collected from 100 peri-urban farming households in Faisalabad using structured questionnaires. The survey included variables such as land ownership, crop types, agricultural income, and vegetable consumption. Additionally, land use changes from 2018 to 2023 were analyzed using GIS tools to observe the impact of urban expansion. Descriptive statistics and Chi-Square tests were applied to assess relationships between food access, nutritional perceptions, and consumption patterns.Results and Discussion: Findings revealed that households allocated an average of 12.9 Kanals for agriculture, growing seasonal vegetables like turnip, carrot, spinach, and peas. Nearly half of the produce was consumed domestically, while the remainder was sold locally. A significant association (p < 0.05) was found between positive nutritional perceptions and regular access to fresh produce. However, limited government support, weak stakeholder coordination, and inadequate market access emerged as key barriers.Conclusion: Peri-urban agriculture significantly contributes to household nutrition and food access in Faisalabad. Yet, its broader impact is limited by institutional gaps. Strengthening collaboration among farmers, policymakers, extension workers, and markets is essential for making peri-urban agriculture more resilient and sustainable in urban Pakistan
ECG Lead Selection for Disease Diagnostics Using CNN-Transformer
Get PDFElectrocardiography (ECG) is crucial for diagnosing cardiovascular diseases (CVDs), which cause millions of deaths each year. This study addresses the challenge of CVD diagnosis in rural areas, where there is a shortage of skilled healthcare professionals and medical equipment. This study proposes a novel method to systematically compare different ECG leads using Deep Learning techniques, specifically a 1D CNN Transformer, to detect anomalies from minimal disturbances. The analysis was conducted using the PTB-XL dataset and further validated with Holter ECG-based records from the St. Petersburg INCART database. Minimal pre-processing was applied, limited to baseline wander removal, to maintain the intrinsic information of each lead. The results indicate that utilizing all leads significantly improves the F1 score, although lead II, V1, and V2 also provide comparable results in the INCART database. This study demonstrates that fewer leads can be effectively used to diagnose diseases, facilitating the creation of low-cost ECG machines suitable for deployment in rural areas. The code is publicly available at https://github.com/nabeelraza-7/ecg-lead-selection
Land Degradation Risk Assessment in District Dir, Pakistan
Get PDFSoil erosion is a global concern, influenced by terrain, vegetation, soil, and climate factors. Traditionally, field-based techniques have been utilized for the measurement of soil erosion. In the present study, Remote Sensing and Geographic Information System (RS/GIS) techniques are used for soil erosion estimation. The Revised Universal Soil Loss Equation (RUSLE) is frequently utilized, incorporating various elements such as soil erodibility, rainfall erosivity, slope steepness, Land Use and Land Cover (LULC), and conservation practices. This study focuses on the Dir district in Pakistan, integrating the RUSLE model with RS and GIS to identify soil erosion-prone areas. The goal is to implement targeted interventions and sustainable land management practices to mitigate soil erosion in these areas. The output of the RUSLE model identifies key zones that need to be addressed to prevent further land degradation. This study also indicates higher C-factor values in Upper and Lower Dir, ranging from 0.001 to 0.2. Soil loss was calculated using all factors (R, K, LS, CP), showing that soil loss is approximately 31.6 tons/ha/yr in Upper Dir and 22.88 tons/ha/yr in Lower Dir, which is higher in Upper Dir due to high elevation (>30m) and more rainfall in Upper Dir (1275mm). Furthermore, annual rainfall values ranging from 508 mm to 1275 mm were noted, resulting in maximum rainfall erosivity values of 572.87 MJ mm ha/h/year in Upper Dir and 568.16 MJ mm ha/h/year in Lower Dir. Thus, this study provides critical data for society and policymakers to implement targeted soil conservation measures and sustainable land management systems, thereby mitigating soil erosion and preventing further land degradation in the district of Dir