Innovative Computing Review

oGoogleNet: An Optimized GoogleNet for Chest Infection Detection on the COVID-19 Dataset

2025-12-20T20:23:04+00:00

The outbreak of SARS and, more recently, COVID-19 has highlighted the
need for accurate and quick diagnosis of chest diseases for pandemic prevention. While the
handling of theCOVID-19 pandemic has drawn attention to the weaknessesin the
healthcare systems worldwide, it has also enabled us to fully utilize the massive amounts
of data at our disposal in order to devise strategies for better handling outbreaks in the
future. Chest infection is a crucial symptom used to diagnose COVID-19 cases.
Moreover, it may also lead to various other diseases, including pneumonia, asthma, and
bronchitis. Researchers havebeen working on automatic chest infection detection for the
last few decades. In this study, we present oGoogleNet, a deep learning architecture for
chest infection detection, developed by optimizing GoogleNet through the addition of
layers and the modification of activation functions. The oGoogleNet iscompared with the
existing state-of-the-art deep networks on eight standard chest infection datasets,
containing 12,389 radiographs (with 777 COVID-19 radiographs). The experiments
demonstrate that oGoogleNet outperforms the other systems and achieves an accuracy of
91.25%.

A Load Classification Strategy using NILM and DNN for Potential Demand-Side Management

2025-12-20T20:16:34+00:00

Uncoordinated and unplanned increase in electricity demand is a critical concern in recent times owing to increasing population and appliance utilization. Significant focus has been given on optimizing load patterns for appliances and capitalizing the potential for savings in domestic energy management. This paper develops a low-cost demand-side management system for residential application through smart energy meters, combined with non-Intrusive load monitoring (NILM) and machine learning for accurate load disaggregation. This paper presents a real-time consumption-based dynamic pricing algorithm, exploiting the use of deep neural networks (DNN) for the classification of essential and non-essential loads with the help of real-time collected datasets. The system provides live energy monitoring through Modbus RTU, RS485 protocols, and a Postgre SQL database, which provides data visualization on a Power BI dashboard highlighting real-time advice on optimization of the consumed energy. The proposed approach demonstrates an effective demand response (DR) mechanism, shifting electricity consumption to off-peak hours throughout out the day without reducing overall energy use, hence optimizing the overall load curve metrics and enhancing energy efficiency.

ADVANCED CYBERSECURITY: DETECTION OF ANOMALIES AND CYBER ATTACKS USING HYBRID MACHINE LEARNING MODEL

2025-12-22T10:52:48+00:00

Automated systems can now identify different forms of anomalies in network traffic patterns and threats simultaneously because of the sophisticated techniques employed in modern cyber security systems. This research work devised an intelligent detection method using Long Short-Term Memory (LSTM) and the efficacious machine learning extreme gradient boosting (XGBoost) algorithm to enhance cyber threat detection accuracy. Using the synthetic minority over-sampling technique (SMOTE), the model enhances its performance by creating additional synthetic minority data points, thus, balancing the dataset and reducing bias. The model learns to capture highly complex non-linear relationships in the data which improves overall performance across different attack scenarios. The model design was tested with real network traffic and was found to have an impressive 98% accuracy. The obtained accuracy of this solution demonstrates its value in real world applications of cybersecurity since it enables the rapid identification of zero day and advanced persistent threats among many other cyber-attacks without loss of precision in the process. Likewise, our proposed approach also addresses the data imbalance issues and improves the model’s ability to accurately and sensitively detect anomalies.

SpecX – a Linux software suite for Electrical and Computer Engineering Education

2025-12-22T10:54:11+00:00

The Undergraduate program in Electrical and Computer Engineering (ECE) requires extensive use of tools within its 3 to 4-year curriculum. These tools are accessible to students at the University, where students spend several hours completing their assignments, which is often hard. Hence, students (sadly) often resort to either using pirated software or copying their colleagues' work. To facilitate experimental learning, the authors have coupled Linux-based tools free for academic use, all together as an easy-to-use, GUI package that enables an automatic installation and configuration of 207 tools catering to the entire ECE undergraduate program. The developed solution, SpecX, provides instant relief to students and teachers engaged in ECE education globally.

AI-Driven Market Forecasting in Emerging Economies: A Systematic Review of GANs and Web Scraping Techniques for the Pakistan Stock Exchange

2025-12-22T10:57:30+00:00

Evaluated on PSX-100 index data (2018–2023), LSTM models and ARIMA have produced less accurate and dependable results compared to hybrid model, achieving a 4.80% MAPE (22.4% improvement over ARIMA) and 77.6% directional accuracy. Model interpret- ability was enhanced using SHAP (Shapley Additive Explanations), providing actionable insights for investors. Ethical web scraping ensured compliance with data governance standards. The framework’s adaptability to other emerging markets (e.g., Bangladesh, Nigeria) is demonstrated, alongside policy recommendations for AI adoption in Pakistan’s financial sector. Predicting stock market trends in emerging economies has always been difficult due to unstable conditions and limited data availability. Although artificial intelligence (AI) is rapidly transforming financial forecasting worldwide, its use in the Pakistan Stock Exchange (PSX) is still quite rare. This research attempts to fill that gap by developing a new approach that blends traditional Generative Adversarial Network (GAN)-based technical analysis with modern AI tools and real-time data scraping. Instead of relying solely on old or static market indicators, the proposed model collects fresh PSX data like prices, volumes, and investor sentiment from multiple online sources and uses a deep learning technique (LSTM) to process it alongside GAN's geometric patterns. The results of the evaluation of the hybrid forecasting approach using PSX-100 index data from 2018 to 2023 were encouraging and the findings which we got was so amazing It provided more accurate and reliable forecast trends and fewer prediction errors than popular prediction techniques like ARIMA and simple GAN models. This models clarity is one of its best features; analysts and investors may easily understand the logic and reasoning behind its results by using a SHAP-based interpretation. Crucially, the information was obtained through the use of ethical web scraping methods, promising that the procedure sticked to responsible data standards. This article bridges the gap between traditional analysis and explainable AI, delivering a robust, rules - based yet adaptive forecasting tool for Pakistan Stock Exchange Traders.