Kathmandu College of Management

The increasing prevalence of lifestyle diseases has raised the demand for personalized dietary counseling. Traditional diet planning was much reliant on expert recommendations and failed to meet specific requirements. This study developed a Diet Recommendation System using K-Means clustering to provide personalized diet recommendations based on age, weight, height, BMI, activity level, body type, and health goals. Across every meal, the K-Means clustering algorithm produced stable and optimal results based on the three different metrics used to evaluate clustering performance - barycenter, silhouette width, and clustering validity metric. For breakfast, K-Means produced the best clustering solution ($k=3$, Silhouette $=0.7879$, Davies-Bouldin $=0.3731$, Calinski-Harabasz $=403$), the second-best clustering solution ($\mathrm{k}=2$, Silhouette $=0.7342$, Davies-Bouldin $=$ 0.4592, Calinski-Harabasz $=399$), and the lowest performing clustering solution ($k=4$, Silhouette $=0.6715$, Davies-Bouldin $=$ 0.4831, Calinski-Harabasz $=656$). It has a better clustering solution than DBSCAN for breakfast, but DBSCAN could not provide sufficient clusters for lunch or dinner. The GMM clustering algorithm for breakfast was comparable to K-Means based on silhouette width and Davies-Bouldin metric (Silhouette $=0.7879$, Davies-Bouldin $=0.3731$). GMM had average results for lunch (Silhouette $=0.7159$, Davies-Bouldin $=0.5294$), and GMM produced the lowest clustering results for dinner (Silhouette $=0.6614$, Davies-Bouldin $=0.4626$). GMM provided a more complex clustering solution (higher AIC and BIC) than KMeans. Overall, K-Means offers the best balanced trade-off between quality, stability, and interpretability for the mealrecommendation system proposed in this research. Results were reflective of success in supporting weight management, muscle gain, and overall health. The merging of nutrition science and machine learning enabled scalable, accurate, and dynamic diet suggestions that provided a viable solution to a healthier lifestyle.

Cloud computing is highly integrated with modern digital infrastructures, expansion of computing, and its services being flexible and adaptable. However, with the increase of cyber threats, the challenge of quantum computing and cyber threats remains unprecedented. The potential of quantum computing remains a complete challenge to establish classical encryption techniques. The difficulty with traditional security techniques is the equilibrium between adaptive shielding and effective computing. The lack of shielding in cloud systems is a complete lack of computational efficiency and bear breaches of cloud systems. This computational efficiency is the gap which this research intends to bridge with Quantum-Resilient Security Optimization for Cloud Architectures (Q-SECURE-OPT). The novel integration with parallel adaptation cloud security, to be discussed in this document, involves the integration of optimization frameworks with post-quantum cryptographic methods. Through the use of parallel processing, adaptive heuristics make latency of real-time intrusion detection, security resource allocation, active optimization of key distribution, and intrusion deqetection adaptable to a virtually 0 gradient of a target function parallel closure. This research demonstrated that post-quantum cryptographic methods enhance encryption efficiency to the extent of 35% and analytically withstand quantum attacks directed through adaptive heuristics. The results indicate that Q-SECURE-OPT not only strengthens the security posture of cloud infrastructures but also ensures scalability, responsiveness, and robust adaptive performance under diverse workload conditions.

Purpose/Aim: This study evaluates compliance with the Nepal National Building Code (NNBC) and municipal bylaws in residential construction in Godawari Municipality, Lalitpur District, Nepal, to identify gaps and propose measures for enhancing earthquake-resistant construction. Design/Methodology/Approach: A qualitative approach was used, combining literature review, field surveys of 147 reinforced concrete (RC) buildings (78 completed, 69 under construction), questionnaires with 32 contractors/masons, and key informant interviews (KII) with municipal officials and consultants. Compliance was assessed using checklists tailored for completed and under-construction buildings, with data analyzed descriptively in MS Excel. Findings: Only 27 of 78 completed buildings fully complied with approved designs, with deviations in staircase dimensions, beam depth, and plinth area. Under-construction buildings showed high compliance in structural elements (e.g., 100% in columns, 98% in slabs) but significant deficiencies in concrete mix proportions (24% compliance) and rebar lapping (32% non-compliant in columns). Masons exhibited limited awareness of building code provisions (25%) despite better knowledge of earthquake phenomena (93.75%). Non-compliance stemmed from low awareness, inadequate training, and owner-driven design changes. Originality/Value: The study provides localized insights into building code compliance in Godawari Municipality, comparing findings with Sisne and Birendranagar Municipalities, and offers practical recommendations for improving enforcement, training, and awareness to enhance seismic resilience.

Rapid expansion of the Internet of Things (IoT) and mobile applications has increased the demand for low latency and efficient computational resource management. Although offering high processing power, cloud computing suffers from network congestion and latency issues, making edge computing a viable alternative. Task offloading in cloud-edge environments is crucial for optimizing resource allocation, reducing delays, and enhancing system performance. For intelligent job offloading in an Edge-Cloud architecture, this study suggests a hybrid reinforcement learning model based on Deep Deterministic Policy Gradient (DDPG) and Proximal Policy Optimization (PPO). The model is trained on the datasets simulated in EdgeCloudSim simulator and consisting of tasks, network conditions, computational capabilities of servers, and energy efficiency constraints. The results of the experiment shows that the suggested model considerably reduces task execution time, lowers energy usage, and boosts system efficiency. This work provides a robust deep reinforcement learning-based solution for optimizing task offloading in future edge-cloud computing infrastructures.

Deepfakes and synthetic media, powered by AI technologies like Generative Adversarial Networks (GANs) and Diffusion Models, pose significant threats to digital trust and security. These tools enable the creation of highly realistic forgeries, which are increasingly exploited in social engineering attacks to spread misinformation, manipulate emotions, and impersonate trusted individuals. This chapter explores the role of deepfakes in enhancing social engineering tactics, such as phishing, CEO fraud, and disinformation campaigns. It reviews current detection methods, including machine learning techniques to identify inconsistencies in synthetic media, and discusses the ethical, societal, and policy challenges posed by this emerging threat. By examining the intersection of deepfakes and social engineering, this chapter highlights the need for robust detection technologies, public awareness, and regulatory measures to address the growing risks of AI-generated content in cybersecurity.

Community pharmacies in Nepal serve as vital access points for healthcare services and medication. However, compliance with Good Pharmacy Practice (GPP) standards remains suboptimal despite the existence of national guidelines and regulatory frameworks. This perspective highlights the need for GPP implementation, identifies key challenges, and outlines the potential opportunities associated with effective compliance. Major barriers include limited awareness of GPP standards, weak regulatory enforcement, unauthorized personnel, financial constraints, and public misconceptions about the pharmacists role. Additionally, factors such as high workloads, inadequate infrastructure, and lack of incentives further hinder compliance. Nonetheless, GPP adoption offers significant benefits: promoting rational drug use, enhancing patient safety, and strengthening pharmacists role in healthcare delivery. It also promotes transparency, professional development, and alignment with global pharmacy standards. Strengthening regulatory oversight, advancing professional education, and raising public awareness are essential to overcoming current barriers. Effective GPP implementation can standardize community pharmacy services and elevate Nepal’s healthcare system to meet international quality benchmarks.

The rapid growth of internet-based activities has led to an increase in cyber threats, particularly in the form of malicious URLs. Detecting these URLs is crucial for user safety. As the source of URL sharing can be in any form of digital media to popular social networks like Facebook, WhatsApp, Viber, Instagram. The research thoroughly explores the use of machine learning techniques such as Random Forest, Support Vector Machines, and Artificial Neural Networks for detecting these malicious URLs. This study employs a Kaggle dataset containing labeled URLs, focusing on extracting relevant features using only lexical-based features. Experimental results show promising outcomes for all three algorithms, indicating their ability in identifying and categorizing malicious URLs. In particular, the ANN algorithm outperforms SVM and RF by achieving an accuracy of $98.97 \%$ in detection. This study aids in enhancing the security of digital systems and protecting individuals from potential cyber dangers associated with harmful URLs.

Monoamine oxidase B (MAO-B) serves as a critical target in the management of neurodegenerative diseases (NDDs) such as Alzheimer’s and Parkinson’s due to its role in regulating oxidative stress and dopamine metabolism. In this context, phytochemicals from Oxalis species, known for their neuroprotective properties, were explored for their potential MAO-B inhibitory activity using computational approach. Plant-derived compounds, offering a better safety profile than synthetic drugs and greater cost-effectiveness, present a promising avenue for developing alternative therapeutic strategies. Molecular docking (MD), molecular dynamics simulations (MDS), and binding free energy calculations were employed to evaluate the inhibitory potential of Oxalis phytochemicals against MAO-B (PDB ID: 4A79). Stable ligand-protein complexes with optimal docking scores were selected, and key parameters from molecular dynamics trajectories, including binding stability and interactions, were analyzed to identify high potential inhibitors of MAO-B for therapeutic development. Results showed beta-sitosterol (−11.92 kcal/mol), squalene (−11.89 kcal/mol), etretinate (−11.46 kcal/mol), rhoifolin (−11.44 kcal/mol), and swertisin (−11.13 kcal/mol) demonstrated superior binding affinities compared to the native ligand (−11.12 kcal/mol). Three additional compounds; phloridzin (−11.10 kcal/mol), rhapontin (−11.02 kcal/mol), and diosmetin 7-O-beta-D-glucopyranoside (−10.96 kcal/mol) exhibited better binding than reference drugs. The predominant interactions between protein and ligand were hydrophobic, with hydrogen bonds and Pi-stacking enhancing the complexes’ stability. The evaluation based on geometrical and thermodynamic metrics derived from 200 ns MDS, identified rhoifolin, beta-sitosterol, and swertisin as promising MAO-B inhibitors. Minimal translational and rotational movements of these ligands within the catalytic site of MAO-B under quasi-physiological conditions suggested effective inhibition. Preserved thermodynamic feasibility reinforced these findings. ADMET analysis identified squalene and beta-sitosterol as CNS active candidates with favorable pharmacokinetics, while etretinate, rhoifolin, and swertisin may act as peripheral modulators, requiring optimization for improved CNS delivery. Further experimental validation of efficacy, pharmacokinetics, and safety is recommended to advance the therapeutic potential of these hit candidates.

The research aimed to assess the impact of financial risk on the performance of cooperatives in Nepal, focusing on credit, operational, and market risks are independent variables and financial performance is the dependent variable. The Research data was collected from financial statements and regulatory records spanning mid-July, 2015 to mid-July, 2024, though the dataset was unbalanced due to variations in the licensing years of the cooperatives. Descriptive statistics, including mean, standard deviation, and range, summarized the data, while correlation coefficient analysis examined the strength and direction of relationships between variables. Regression analysis explored the causal impact of financial risks on cooperative performance. The findings revealed that credit risk had a marginal but its positive effect on financial performance, operational risk had a significant and positive influence, and market risk demonstrated a substantial positive impact. The study underscores the importance of comprehensive risk management practices and recommends that cooperatives implement strategies addressing all dimensions of financial risk, including the recruitment of skilled risk management professionals, to enhance overall performance.

The integration of AI with Building Information Modeling (BIM) and Digital Twin technology is reshaping the architectural landscape, bridging the gap between cutting-edge construction practices and the preservation of cultural heritage. This article explores how AI-driven BIM streamlines construction through predictive maintenance, cost optimization, and real-time project adjustments, enabling the creation of efficient, resilient, and sustainable structures. Simultaneously, digital twin technology, powered by AI, offers new avenues for historic preservation by producing data-rich, 3D models that allow for precise monitoring and virtual restoration of heritage sites. By examining real-world applications in both modern buildings and historic landmarks, this article demonstrates how AI’s role in architecture not only advances building practices but also safeguards architectural history for future generations.