International Journal of Systematic Innovation

A systematic approach to corporate innovation excellence

Koray Altun — 2025-04-08

While international standards on innovation management have gained interest, “excellence” in innovation management has not been thoroughly studied in the literature. To address this gap, this study proposes the “Innovation Excellence Model” for corporate innovation. This approach aims to provide a concise way of excellence in corporate innovation system design. This model focuses on three important components of the system: innovation execution system, innovation organization, and innovation engine. This model is based on three different innovation engines (idea-driven, analysis-driven, and research-driven) and proposes a card-based control system to balance workload and project flows. The integration of card-based control and its simulated case provides a tangible and effective means of translating theoretical concepts into practical execution. A novel key performance indicator, “CIP – Corporate Innovation Performance” is also introduced for monitoring the excellence degree. By fostering a holistic understanding of excellence in corporate innovation, the model enables organizations to navigate the design of innovation management system, propelling them toward excellence and growth.

Structure learning of Bayesian networks using sparrow optimization algorithm

Shahab Wahhab Kareem — 2025-04-11

Bayesian networks are powerful analytical models in machine learning, used to represent probabilistic relationships among variables and create learning structures. These networks are made up of parameters that show conditional probabilities and a structure that shows how random variables interact with each other. The structure is shown by a directed acyclic graph. Despite the NP-hard nature of learning Bayesian network structures, there has been significant progress in improving the accuracy of approximation solutions. The main focus is on score-based search strategies, which make use of functions to evaluate network models and identify structures with high scores. This study is significantly focused on structure learning Bayesian networks using the Bayesian Dirichlet equivalent uniform scoring function and metaheuristic search strategies. To this end, this paper presents the sparrow optimization algorithm (SOA), a new metaheuristic algorithm derived from the foraging behavior of sparrows. SOA performs a concurrent optimization in the solution space by simultaneously performing a local and global search that leads to the discovery of near-optimal structures. The results from our experiments on several benchmark datasets show that SOA yields overall better performance than SA and greedy search algorithms. In particular, it is claimed that by applying the proposed approach of SOA, the convergence speed is significantly higher compared with the existing ones; F1 score is 0.35 and 0.05 for the Hamming distance with better results. Given these results, signed operators prove to be very efficient in SOA’s Bayesian network structure learning as a concept, especially for real-world use.

Secure mobile cloud data using federated learning and blockchain technology

G. Matheen Fathima — 2025-04-08

In the current era, mobile cloud (MC) transactions raise concerns over the data stored in the MC. These data can be tampered with by third parties, leading to data loss and information misplacement. Such security breaches can be mitigated by implementing federated learning (FL). FL refers to a distributed data learning approach that trains data without revealing the information to the server or coordinator. It uses the current model data for training and then sends the updated model to the coordinator or server. The server collects the updated trained models from all clients and aggregates them into a single global model. This updated model is then communicated back to the clients. FL, when implemented with MC, protects user privacy, ensures efficient learning, and achieves higher accuracy compared to traditional machine learning algorithms. We propose the implementation of MC FL using blockchain, a model designed to protect user data by maintaining it on edge devices and sending the updated model to the server after training. Finally, the data-generated model will be stored in the blockchain network, preventing data tampering and providing a higher level of security and privacy for the data.

A comparative study of traditional machine learning models and the KNN-KFSC method for optimizing anomaly detection in VANETs

Ravikumar Ch — 2025-04-08

In this research, we conducted a comparative analysis of traditional machine learning techniques and the innovative K-nearest neighbors-K-fuzzy subspace clustering (KNN-KFSC) methodology to detect anomalies in vehicular ad hoc network (VANET) infrastructures. Our evaluation included models such as support vector machine (SVM), random forest (RF), logistic regression (LR), and KNN. The KNN-KFSC model demonstrated exceptional performance with an overall accuracy rate of 99% in handling densely contextual data. It consistently exhibited high accuracy, recall, and F1 score metrics, indicating its effectiveness in detecting a broad spectrum of anomalies across various types of attacks in VANETs. In contrast, the RF algorithm achieved an 89% accuracy rate, showcasing competency in specific domains but revealing limitations in others. Both LR and SVM models exhibited identical accuracy rates of 92%. While effective in identifying specific types of attackers, these models showed weaknesses, potentially due to overfitting or inadequate management of dataset complexity. The KNN-KFSC approach emerged as the most promising option for detecting anomalies in software-defined VANETs, evidenced by its superior performance in accuracy and precision. Our findings underscore the necessity of advanced intrusion detection system techniques and highlight the importance of model refinement to address data imbalances and improve anomaly detection in VANET systems.

Hybrid prediction model by integrating machine learning techniques with MLOps

Poonam Narang — 2025-04-08

Recent advancements in machine learning (ML) have sparked widespread interest in integrating DevOps capabilities into software and services within the information technology sector. This objective has compelled organizations to revise their development processes. We propose a ML operations model based on meta-ensembling algorithm for gradient boosting regressor with a case study of real estate price prediction. The train and test dataset is loaded with (1460,80) predictive variables, with the sale price as the target variable. The forecasting model is developed using an artificial neural network and a linear logistic regression model, such as LASSO, alongside with the Heroku tool for model deployment. The methodology addresses different steps of data pre-processing, and feature engineering, followed by feature selection, model building, evolution, creating, and calling application programming interfaces for deployment as IaaS, under research, development, and production environment phases. The model is built using the Anaconda Jupyter notebook with various Python libraries and Docker to ensure reproducibility and robustness. To ensure good business value, the performance of the proposed and implemented model is evaluated using different classification metrics, such as area under the curve-ROC for correct assessment measure, alongside accuracy metrics like mean squared error, root mean squared error, and R-squared. Our work serves as a useful reference for building and deploying ML pipeline platforms in practice.

Measuring the accuracy of time series reduction methods based on modified dynamic time warping distance calculations

Anupama Jawale — 2025-04-08

Representation of sensor data in the form of time series is a crucial aspect of numerous related tasks such as comparison, reduction, clustering, and classification. Time series representation methods included in most programming languages/integrated development environments support dimensionality reduction, data preprocessing, and feature extraction for time series data, as do several normalization techniques. This research study focused on 14 different methods of dimensionality reduction from the TSepr (R Studio) package on eight different time series, which are collections of sensor data of varying lengths. The similarity of the reduced time series and the original time series is compared using a modified version of dynamic time warping with time alignment measurement. These methods are further combined with the Gaussian kernel function to normalize the distance between variously aligned series. The results showed that perceptually important points (PIP) and piecewise linear approximation (PLA) were found as the best methods for TS reduction with a minimum deviation (error term) as low as 5 – 12%. The results also indicate that PIP performs significantly differently compared to seasonal decomposition, while there are no significant differences between PIP and the other methods (PLA, FEACLIPTREND, and FEACLIP). In addition, this research study demonstrated the development of an interactive web-based application in which time series are stored in csv files, and the distance between them is calculated through the chosen reduction method.

Optimizing cloud-based intrusion detection systems through hybrid data sampling and feature selection for enhanced anomaly detection

Sadargari Viharika — 2025-04-08

To enhance detection accuracy in network intrusion scenarios, this study proposes an optimized intrusion detection system (IDS) framework that integrates advanced data sampling, feature selection, and anomaly detection techniques. Leveraging random forest (RF) and genetic algorithm, the framework optimizes sampling ratios and identifies critical features. In contrast, the isolation forest algorithm detects and excludes outliers, refining dataset quality and classification performance. Evaluated on the UNSW-NB15 dataset, comprising over 2.5 million records and 42 diverse features, the proposed framework demonstrates significant improvements in anomaly detection, including reduced false alarm rates and enhanced identification of rare threats, such as shellcode, worms, and backdoors. Experimental results reveal that the RF-based model achieves an F1 score of 91.8% and an area under the curve (AUC) of 96%, outperforming traditional machine learning models and standalone RF classifiers. The integration of extreme gradient boosting (XGB) and its optimized variant, XGBGA, further enhances the framework, with XGBGA achieving the highest performance metrics, including an F1 score of 92.8% and an AUC of 97%. These findings underscore the importance of data optimization strategies in improving the accuracy and reliability of IDSs, particularly in handling imbalanced datasets and diverse network traffic. Future work will focus on real-time processing capabilities to handle streaming data and expanding the framework’s applicability to domains such as fraud detection and cybersecurity, where precise anomaly detection is essential.

YOLOXpress: A lightweight real-time unmanned aerial vehicle detection algorithm

Nguyen Ngoc Hung — 2025-04-08

The widespread use of drones has made drone detection a critical factor in various fields, particularly in security and defense. However, this task presents unique challenges due to the high speed, small size, and ability of drones to blend into their surroundings, which can hinder detection effectiveness. This paper introduces enhancements to the You Only Look Once (YOLO)-v8 model to improve real-time drone detection capabilities, especially when deployed on resource-constrained devices. We propose an improved model called YOLOXpress, which optimizes both processing speed and model size while maintaining an acceptable level of accuracy. By replacing the Cross-Stage Feature Fusion modules in the Backbone and Neck with Re-parameterization Convolution and RepC3 modules, we significantly reduced the number of computations, achieving a 12.25% increase in processing speed (frames per second) and a 69.96% reduction in model size. Although there was a 6% decrease in average accuracy compared to the original YOLO-v8 model, YOLOXpress remained effective for real-time drone detection. Experiments conducted on the TIB-Net dataset confirmed that this model is highly suitable for deployment on resource-limited devices, such as compact embedded systems.

A blockchain-based solution to combating identity crime and credit card application fraud using data mining algorithms

Amol Jagdish Shakadwipi — 2025-04-08

Fraud, specifically identity theft and credit card fraud, poses significant threats not only to financial institutions but also to their users. In response to this growing problem, we present an innovative approach that integrates self-sovereign identity management based on blockchain and complex data analysis. Our comprehensive solution is designed to revolutionize identity verification in credit card application processes by significantly enhancing security and reducing vulnerability to identity fraud. The system that will be developed from our solution will help users obtain self-sovereign identity credentials through blockchain technology or distributed ledger technology, granting them full control over their personal data. This approach has been proven to drastically reduce the likelihood of identity theft, and it does not require centralization of data. Besides, the use of blockchain technology ensures more credible records of identification, as they are transparent and immutable. At P&L, we combine smart data mining with blockchain-based identity solutions as our primary strategy. These algorithms detect patterns and anomalies related to identity theft in massive datasets. The technology can quickly flag suspicious activity and verify identity claims in real-time by continuously comparing recent user activity with historical data.