Akbar Ghobakhlou
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Akbar Ghobakhlou has a total of 25 co-authors. Here are the top co-authors by number of shared publications:
- Ajit Narayanan
School of Engineering, Computer and Mathematical Sciences, Auckland University of Technology, Auckland 1010, New Zealand
6 shared publications - Rujia Chen
Auckland University of Technology
3 shared publications - Bunleng Chhun
MJ Realties Limited, Auckland 1072, New Zealand
2 shared publications - Chengran Li
School of Integrated Circuits, BNRist, Tsinghua University, Beijing, China
2 shared publications - Jairo Gutierrez
School of Engineering, Computer and Mathematical Sciences, Auckland University of Technology, Auckland 1010, New Zealand
2 shared publications - Minh Nguyen
Computer and Information Sciences Department, Auckland University of Technology, Auckland 1010, New Zealand
2 shared publications
Latest Publications
- Open Access
- 19 November 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (22) , 11494
Evaluating Deep Multimodal Learning for Forecasting Review-Based Short-Term Rental Platform Activity from Digital Listing Content and Platform Signals: Evidence from New Zealand
Digital short-term-rental platforms combine structured property information, historical engagement, platform-management information, and host-written text, yet it is unclear whether multimodal gains persist in later periods. This study develops a privacy-aware longitudinal forecasting framework using five nationwide New Zealand Inside Airbnb snapshots (October 2025 to August 2026). Listings are matched across time, and cumulative review growth is standardised to a 90-day horizon as a review-based proxy for platform activity. Two windows (W1, W2) are used for development and a later window (W3; ) is the held-out temporal test. Five models, from a recent-activity persistence baseline to deep multimodal fusion, are compared. The persistence baseline achieved the lowest W3 MAE (1.974). Machine-learning fusion was the strongest learned model under the primary specification (MAE 2.476, , Spearman ), but only narrowly ahead of structured XGBoost, and removing text improved W3 error, so the gain cannot be attributed to text. Historical engagement dominated SHAP importance and ablation, deep fusion did not improve temporal generalisation, and eight of 33 regions had negative . The contribution is methodological: model-level gain, modality-specific value, and temporal robustness are evaluated separately. The target is not interpreted as bookings, occupancy, demand, or revenue.- 19 November 2026
- Published by MDPI AG in Applied Sciences
- Vol. 16 (22) , 11494
Open AccessEvaluating Deep Multimodal Learning for Forecasting Review-Based Short-Term Rental Platform Activity from Digital Listing Content and Platform Signals: Evidence from New Zealand
Digital short-term-rental platforms combine structured property information, historical engagement, platform-management information, and host-written text, yet it is unclear whether multimodal gains persist in later periods. This study develops a privacy-aware longitudinal forecasting framework using five nationwide New Zealand Inside Airbnb snapshots (October 2025 to August 2026). Listings are matched across time, and cumulative review growth is standardised to a 90-day horizon as a review-based proxy for platform activity. Two windows (W1, W2) are used for development and a later window (W3; ) is the held-out temporal test. Five models, from a recent-activity persistence baseline to deep multimodal fusion, are compared. The persistence baseline achieved the lowest W3 MAE (1.974). Machine-learning fusion was the strongest learned model under the primary specification (MAE 2.476, , Spearman ), but only narrowly ahead of structured XGBoost, and removing text improved W3 error, so the gain cannot be attributed to text. Historical engagement dominated SHAP importance and ablation, deep fusion did not improve temporal generalisation, and eight of 33 regions had negative . The contribution is methodological: model-level gain, modality-specific value, and temporal robustness are evaluated separately. The target is not interpreted as bookings, occupancy, demand, or revenue. - Open Access
A Multi-Source Data Integration and Temporal Evaluation Framework for Short-Term Rental Market Screening: Evidence from New Zealand
Combining short-term rental platform records with official housing statistics requires more than matching locations and dates, because platform records may be collected after their nominal snapshot date, outcomes may be missing, and official statistics are released and revised with a lag. We developed an auditable integration and temporal evaluation framework and applied it to 491,140 listing-origin observations from 11 New Zealand Inside Airbnb snapshots linked to release-aligned building consents, rental bonds and seasonal indicators. The evaluation task was next-snapshot platform disappearance, which is not equivalent to business closure. An audit showed that 98.48% of origins were scraped after their nominal snapshot date and that two horizon-dependent predictors were not available prospectively. On a locked final transition (51,096 listings), the frozen XGBoost model reached a ROC-AUC of 0.7432, and the 10% of listings with the highest predicted risk contained 31.65% of observed disappearances (precision 7.10%). Calendar configuration and review history gave the largest ranking gains. Building-consent intensity was positively associated with area-level disappearance rates but added little to short-horizon forecasts, and XGBoost did not predict changes in advertised price rank better than a zero-change baseline. Code, checksums and policy-compliant derived data are openly available.
Open AccessA Multi-Source Data Integration and Temporal Evaluation Framework for Short-Term Rental Market Screening: Evidence from New Zealand
Combining short-term rental platform records with official housing statistics requires more than matching locations and dates, because platform records may be collected after their nominal snapshot date, outcomes may be missing, and official statistics are released and revised with a lag. We developed an auditable integration and temporal evaluation framework and applied it to 491,140 listing-origin observations from 11 New Zealand Inside Airbnb snapshots linked to release-aligned building consents, rental bonds and seasonal indicators. The evaluation task was next-snapshot platform disappearance, which is not equivalent to business closure. An audit showed that 98.48% of origins were scraped after their nominal snapshot date and that two horizon-dependent predictors were not available prospectively. On a locked final transition (51,096 listings), the frozen XGBoost model reached a ROC-AUC of 0.7432, and the 10% of listings with the highest predicted risk contained 31.65% of observed disappearances (precision 7.10%). Calendar configuration and review history gave the largest ranking gains. Building-consent intensity was positively associated with area-level disappearance rates but added little to short-horizon forecasts, and XGBoost did not predict changes in advertised price rank better than a zero-change baseline. Code, checksums and policy-compliant derived data are openly available.
- Open Access
- 28 July 2026
- Published by New Zealand Ecological Society in New Zealand Journal of Ecology
- Vol. 50, 3640
Audible and ultrasonic vocalisations of two captive stoats (Mustela erminea): structural categorisation and an open annotated dataset
Stoat (Mustela erminea) vocalisations remain sparsely characterised, despite the species' significance as a major invasive predator in Aotearoa | New Zealand. Here, we describe the audible and ultrasonic vocalisations recorded from an adult male-female pair held in laboratory enclosures over three days, with blank-room control recordings. Audible calls attributed to the stoats were categorised into chitter, grunt, and hiss; the hiss is a narrowband c. 10 kHz tone of several seconds’ duration, consistent with early descriptive accounts.
- 28 July 2026
- Published by New Zealand Ecological Society in New Zealand Journal of Ecology
- Vol. 50, 3640
Open AccessAudible and ultrasonic vocalisations of two captive stoats (Mustela erminea): structural categorisation and an open annotated dataset
Stoat (Mustela erminea) vocalisations remain sparsely characterised, despite the species' significance as a major invasive predator in Aotearoa | New Zealand. Here, we describe the audible and ultrasonic vocalisations recorded from an adult male-female pair held in laboratory enclosures over three days, with blank-room control recordings. Audible calls attributed to the stoats were categorised into chitter, grunt, and hiss; the hiss is a narrowband c. 10 kHz tone of several seconds’ duration, consistent with early descriptive accounts.
- Open Access
A Comprehensive Analysis of Security Challenges in ZigBee 3.0 Networks
ZigBee, a wireless technology standard for the Internet of Things (IoT) devices based on IEEE 802.15.4, faces significant security challenges that threaten the confidentiality, integrity, and availability of its networks. Despite using 128-bit Advanced Encryption Standard (AES) with symmetric keys for node authentication and data confidentiality, ZigBee’s design constraints, such as low cost and low power, have allowed security issues to persist. While ZigBee 3.0 introduces enhanced security features such as install codes and trust centre link key updates, there remains a lack of empirical research evaluating their effectiveness in real-world deployments. This research addresses the gap by conducting a comprehensive, hardware-based analysis of ZigBee 3.0 networks using XBee 3 radio modules and ZigBee-compatible devices. We investigate the following three core security issues: (a) the security of symmetric keys, focusing on vulnerabilities that could allow attackers to obtain these keys; (b) the impact of compromised symmetric keys on network confidentiality; and (c) susceptibility to Denial-of-Service (DoS) attacks due to insufficient protection mechanisms. Our experiments simulate realistic attack scenarios under both Centralised and Distributed Security Models to assess the protocol’s resilience. The findings reveal that while ZigBee 3.0 improves upon earlier versions, certain vulnerabilities remain exploitable. We also propose practical security controls and best practices to mitigate these attacks and enhance network security. This work contributes novel insights into the operational security of ZigBee 3.0, offering guidance for secure IoT deployments and advancing the understanding of protocol-level defences in constrained environments.
Open AccessA Comprehensive Analysis of Security Challenges in ZigBee 3.0 Networks
ZigBee, a wireless technology standard for the Internet of Things (IoT) devices based on IEEE 802.15.4, faces significant security challenges that threaten the confidentiality, integrity, and availability of its networks. Despite using 128-bit Advanced Encryption Standard (AES) with symmetric keys for node authentication and data confidentiality, ZigBee’s design constraints, such as low cost and low power, have allowed security issues to persist. While ZigBee 3.0 introduces enhanced security features such as install codes and trust centre link key updates, there remains a lack of empirical research evaluating their effectiveness in real-world deployments. This research addresses the gap by conducting a comprehensive, hardware-based analysis of ZigBee 3.0 networks using XBee 3 radio modules and ZigBee-compatible devices. We investigate the following three core security issues: (a) the security of symmetric keys, focusing on vulnerabilities that could allow attackers to obtain these keys; (b) the impact of compromised symmetric keys on network confidentiality; and (c) susceptibility to Denial-of-Service (DoS) attacks due to insufficient protection mechanisms. Our experiments simulate realistic attack scenarios under both Centralised and Distributed Security Models to assess the protocol’s resilience. The findings reveal that while ZigBee 3.0 improves upon earlier versions, certain vulnerabilities remain exploitable. We also propose practical security controls and best practices to mitigate these attacks and enhance network security. This work contributes novel insights into the operational security of ZigBee 3.0, offering guidance for secure IoT deployments and advancing the understanding of protocol-level defences in constrained environments.
- 26 June 2025
- Published by Springer Nature
- p. 201-214
Comparative Analysis of Different Spectrogram Types for Musical Instrument Classification
- 26 June 2025
- Published by Springer Nature
- p. 201-214
Comparative Analysis of Different Spectrogram Types for Musical Instrument Classification
- 24 June 2025
- Published by Springer Nature
- p. 46-62
ANN-Based Pollution Forecasting Through Short-Term Spatio-Temporal Analysis: A North Island, New Zealand Case Study
- 24 June 2025
- Published by Springer Nature
- p. 46-62
ANN-Based Pollution Forecasting Through Short-Term Spatio-Temporal Analysis: A North Island, New Zealand Case Study
- Open Access
- 22 November 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (23) , 10837
Hierarchical Residual Attention Network for Musical Instrument Recognition Using Scaled Multi-Spectrogram
Musical instrument recognition is a relatively unexplored area of machine learning due to the need to analyze complex spatial–temporal audio features. Traditional methods using individual spectrograms, like STFT, Log-Mel, and MFCC, often miss the full range of features. Here, we propose a hierarchical residual attention network using a scaled combination of multiple spectrograms, including STFT, Log-Mel, MFCC, and CST features (Chroma, Spectral contrast, and Tonnetz), to create a comprehensive sound representation. This model enhances the focus on relevant spectrogram parts through attention mechanisms. Experimental results with the OpenMIC-2018 dataset show significant improvement in classification accuracy, especially with the “Magnified 1/4 Size” configuration. Future work will optimize CST feature scaling, explore advanced attention mechanisms, and apply the model to other audio tasks to assess its generalizability.
- 22 November 2024
- Published by MDPI AG in Applied Sciences
- Vol. 14 (23) , 10837
Open AccessHierarchical Residual Attention Network for Musical Instrument Recognition Using Scaled Multi-Spectrogram
Musical instrument recognition is a relatively unexplored area of machine learning due to the need to analyze complex spatial–temporal audio features. Traditional methods using individual spectrograms, like STFT, Log-Mel, and MFCC, often miss the full range of features. Here, we propose a hierarchical residual attention network using a scaled combination of multiple spectrograms, including STFT, Log-Mel, MFCC, and CST features (Chroma, Spectral contrast, and Tonnetz), to create a comprehensive sound representation. This model enhances the focus on relevant spectrogram parts through attention mechanisms. Experimental results with the OpenMIC-2018 dataset show significant improvement in classification accuracy, especially with the “Magnified 1/4 Size” configuration. Future work will optimize CST feature scaling, explore advanced attention mechanisms, and apply the model to other audio tasks to assess its generalizability.
- 8 October 2024
- Published by Taylor & Francis
- p. 139-161
A comparative study of heuristic inquiry in AI and artistic research
This chapter is written by a computer scientist and an artistic practitioner who embarked on a discussion about the nature of heuristics in their respective fields of inquiry. Although seemingly distant, the authors recognised significant connections between heuristic inquiry in artificial intelligence and in artistic research. The application of heuristics is crucial in navigating the complexities of problem-solving, particularly in fields that demand creative and flexible thinking. The discussion illustrates the use of heuristics in artificial intelligence for tasks such as optimisation and decision-making, and in artistic research for integrating personal intuition and experience into the creative process. By comparing heuristic approaches in these domains, the chapter sheds light on their interdisciplinary potential, revealing how heuristics facilitate innovation and adaptability across diverse landscapes.
- 8 October 2024
- Published by Taylor & Francis
- p. 139-161
A comparative study of heuristic inquiry in AI and artistic research
This chapter is written by a computer scientist and an artistic practitioner who embarked on a discussion about the nature of heuristics in their respective fields of inquiry. Although seemingly distant, the authors recognised significant connections between heuristic inquiry in artificial intelligence and in artistic research. The application of heuristics is crucial in navigating the complexities of problem-solving, particularly in fields that demand creative and flexible thinking. The discussion illustrates the use of heuristics in artificial intelligence for tasks such as optimisation and decision-making, and in artistic research for integrating personal intuition and experience into the creative process. By comparing heuristic approaches in these domains, the chapter sheds light on their interdisciplinary potential, revealing how heuristics facilitate innovation and adaptability across diverse landscapes.
- Open Access
- 27 September 2024
- Published by MDPI AG in Electronics
- Vol. 13 (19) , 3823
A Systematic Review of Deep Learning Techniques for Phishing Email Detection
The landscape of phishing email threats is continually evolving nowadays, making it challenging to combat effectively with traditional methods even with carrier-grade spam filters. Traditional detection mechanisms such as blacklisting, whitelisting, signature-based, and rule-based techniques could not effectively prevent phishing, spear-phishing, and zero-day attacks, as cybercriminals are using sophisticated techniques and trusted email service providers. Consequently, many researchers have recently concentrated on leveraging machine learning (ML) and deep learning (DL) approaches to enhance phishing email detection capabilities with better accuracy. To gain insights into the development of deep learning algorithms in the current research on phishing prevention, this study conducts a systematic literature review (SLR) following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. By synthesizing the 33 selected papers using the SLR approach, this study presents a taxonomy of DL-based phishing detection methods, analyzing their effectiveness, limitations, and future research directions to address current challenges. The study reveals that the adaptability of detection models to new behaviors of phishing emails is the major improvement area. This study aims to add details about deep learning used for security to the body of knowledge, and it discusses future research in phishing detection systems.
- 27 September 2024
- Published by MDPI AG in Electronics
- Vol. 13 (19) , 3823
Open AccessA Systematic Review of Deep Learning Techniques for Phishing Email Detection
The landscape of phishing email threats is continually evolving nowadays, making it challenging to combat effectively with traditional methods even with carrier-grade spam filters. Traditional detection mechanisms such as blacklisting, whitelisting, signature-based, and rule-based techniques could not effectively prevent phishing, spear-phishing, and zero-day attacks, as cybercriminals are using sophisticated techniques and trusted email service providers. Consequently, many researchers have recently concentrated on leveraging machine learning (ML) and deep learning (DL) approaches to enhance phishing email detection capabilities with better accuracy. To gain insights into the development of deep learning algorithms in the current research on phishing prevention, this study conducts a systematic literature review (SLR) following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. By synthesizing the 33 selected papers using the SLR approach, this study presents a taxonomy of DL-based phishing detection methods, analyzing their effectiveness, limitations, and future research directions to address current challenges. The study reveals that the adaptability of detection models to new behaviors of phishing emails is the major improvement area. This study aims to add details about deep learning used for security to the body of knowledge, and it discusses future research in phishing detection systems.