Mahdin Rohmatillah

Mr. Mahdin Rohmatillah

orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
30 August 2019 in Energies
Verified
Journal Article
Open access
PDF file
SPSholeh Hadi Pramono
Sholeh Hadi Pramono
MRMahdin Rohmatillah
Mahdin Rohmatillah
EMEka Maulana
Eka Maulana
RHRini Nur Hasanah
Rini Nur Hasanah
FHFakhriy Hario
Fakhriy Hario
A novel method for short-term load forecasting (STLF) is proposed in this paper. The method utilizes both long and short data sequences which are fed to a wavenet based model that employs dilated causal residual convolutional neural network (CNN) and long short-term memory (LSTM) layer respectively to hourly forecast future load demand. This model is aimed to support the demand response program in hybrid energy systems, especially systems using renewable and fossil sources. In order to prove the generality of our model, two different datasets are used which are the ENTSO-E (European Network of Transmission System Operators for Electricity) dataset and ISO-NE (Independent System Operator New England) dataset. Moreover, two different ways of model testing are conducted. The first is testing with the dataset having identical distribution with validation data, while the second is testing with data having unknown distribution. The result shows that our proposed model outperforms other deep learning-based model in terms of root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In detail, our model achieves RMSE, MAE, and MAPE equal to 203.23, 142.23, and 2.02 for the ENTSO-E testing dataset 1 and 292.07, 196.95 and 3.1 for ENTSO-E dataset 2. Meanwhile, in the ISO-NE dataset, the RMSE, MAE, and MAPE equal to 85.12, 58.96, and 0.4 for ISO-NE testing dataset 1 and 85.31, 62.23, and 0.46 for ISO-NE dataset 2.

9 January 2023 in IEEE/ACM Transactions on Audio, Speech, and Language Processing
Verified
Journal Article
Open access
MRMahdin Rohmatillah
Mahdin Rohmatillah
Jen-Tzung ChienJen-Tzung Chien
Jen-Tzung Chien
Achieving high performance in a multi-domain dialogue system with low computation is undoubtedly challenging. Previous works applying an end-to-end approach have been very successful. However, the computational cost remains a major issue since the large-sized language model using GPT-2 is required. Meanwhile, the optimization for individual components in the dialogue system has not shown promising result, especially for the component of dialogue management due to the complexity of multi-domain state and action representation. To cope with these issues, this paper presents an efficient guidance learning where the imitation learning and the hierarchical reinforcement learning (HRL) with human-in-the-loop are performed to achieve high performance via an inexpensive dialogue agent. The behavior cloning with auxiliary tasks is exploited to identify the important features in latent representation. In particular, the proposed HRL is designed to treat each goal of a dialogue with the corresponding sub-policy so as to provide efficient dialogue policy learning by utilizing the guidance from human through action pruning and action evaluation, as well as the reward obtained from the interaction with the simulated user in the environment. Experimental results on ConvLab-2 framework show that the proposed method achieves state-of-the-art performance in dialogue policy optimization and outperforms the GPT-2 based solutions in end-to-end system evaluation.

4 December 2024 in IEEE Transactions on Audio, Speech and Language Processing
Verified
Journal Article
Open access
Jen-Tzung ChienJen-Tzung Chien
Jen-Tzung Chien
MRMahdin Rohmatillah
Mahdin Rohmatillah
CCChang-Ting Chu
Chang-Ting Chu
Adversarial data augmentation techniques have recently demonstrated potential in enhancing the robustness of machine learning models by identifying potential worst-case data augmentation. However, most of the existing methods implemented a single augmentation strategy for different instances through a process of greedy search, resulting in suboptimal quality of the generated data. Furthermore, previous studies that incorporated reinforcement learning (RL) to apply unique augmentation strategies required high computational cost, as it necessitated a child network to compute the reward during the optimization process. Given these limitations, this study introduces a strategic adversarial data augmentation approach that leverages RL to search for and emulate the worst-case variations through a sequence of augmentation actions. By defining a reward function with an information-theoretic perspective along with the proper definition of state space, a proficient strategy for stacking multiple augmentation strategies can be carried out in an inexpensive way and can be smoothly integrated into classifier training, thereby enhancing model robustness against unseen noises. The proposed adversarial training method was evaluated on ten different types of unseen human-readable noises across six distinct text classification tasks. Experimental results indicate that the proposed method significantly improves model robustness in compensating for unseen noises.

5 August 2025 in Journal of Network and Computer Applications
Verified
Journal Article
Closed access
WYWidhi Yahya
Widhi Yahya
YLYuan-Cheng Lai
Yuan-Cheng Lai
YLYing-Dar Lin
Ying-Dar Lin
MRMahdin Rohmatillah
Mahdin Rohmatillah
BKBinayak Kar
Binayak Kar

14 October 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
ASAdi Saputra
Adi Saputra
ARAbdur Rouf
Abdur Rouf
SMSukir Maryanto
Sukir Maryanto
DSDidik R. Santoso
Didik R. Santoso
Osama EljamalOsama Eljamal
Osama Eljamal
RARafif Amsyar Al Abraar
Rafif Amsyar Al Abraar
Membrane staging and SOFC operating pressure strongly influence the balance between carbon dioxide capture and net energy performance in integrated gasification fuel cell systems. This study investigates the effects of single stage and two stage membrane configurations and solid oxide fuel cell operating pressure on carbon dioxide removal and overall thermodynamic performance using process simulation. A reference configuration without membrane based carbon dioxide capture was compared with single stage and two stage configurations under consistent operating conditions. The analysis considered gross power, auxiliary power, carbon dioxide compression, net power, net lower heating value efficiency, and carbon dioxide removal. The single stage membrane increased gross power from 32,889.98 to 36,055.65 kW and net power from 26,561.20 to 27,639.12 kW, increasing net lower heating value efficiency from 61.11% to 63.59%. The two-stage configuration further increased gross and net power to 38,477.22 and 30,450.49 kW, respectively, with a net lower heating value efficiency of 70.04%. However, carbon dioxide removal decreased from 88.88% to 79.01%, indicating a trade-off between capture and energy performance. Increasing fuel cell operating pressure from 1 to 3 atm increased the average Nernst voltage from 0.886 to 0.922 V and fuel cell power from 11,576.04 to 12,820.26 kW. However, higher auxiliary power consumption reduced net power from 25,969.91 to 25,621.30 kW and net lower heating value efficiency from 59.74% to 58.94%, while carbon dioxide removal remained approximately 88.8%. These results demonstrate the importance of evaluating carbon dioxide capture and integrated net energy performance simultaneously.

26 November 2025 in IEEE Access
Verified
Journal Article
Open access
DSDwi Utari Surya
Dwi Utari Surya
SPSholeh Hadi Pramono
Sholeh Hadi Pramono
PMPanca Mudjirahardjo
Panca Mudjirahardjo
MMMuhammad Aziz Muslim
Muhammad Aziz Muslim
CACries Avian
Cries Avian
MRMahdin Rohmatillah
Mahdin Rohmatillah
Mental stress poses a significant global health burden, with rising prevalence linked to impaired cognitive functioning, reduced productivity, and increased neuropsychiatric risk. Electroencephalography (EEG) provides a non-invasive approach for capturing neural correlates of stress; however, effective classification remains challenging due to the complex spatiotemporal nature of brain activity. To address the limitations of current EEG-based stress classification methods, particularly their limited ability to jointly model temporal, spectral, and spatial dependencies. We propose a novel handcrafted feature-extraction framework, Spatiotemporal Connectivity and Fractal–Spectral Profile (STC-FSP). The framework integrates handcrafted descriptors from four domains, temporal, spectral, fractal, and spatial connectivity (including inter-channel correlation, hemispheric asymmetry, and phase-lag index) to capture both localized neural dynamics and inter-regional synchronization. To enhance cross-domain interaction and reduce redundancy, a log-transformed Pearson correlation is applied across the feature space. The proposed STC-FSP framework was evaluated on the publicly available DASPS dataset using multiple classical machine learning models. In binary classification, K-Nearest Neighbors (KNN) achieved the highest accuracy (93.18%), F1-score (93.89%), and AUC (93.89%), while in multiclass classification, XGBoost yielded competitive accuracy with an exceptionally low inference time of 0.0015 s, demonstrating real-time feasibility. Compared with existing EEG-based stress classification methods such as Baghdadi et al. (2019) (83.5% using Autoencoder), Syakiyla et al. (2023) (89.5% with S-LSMOTE), and Shikha et al. (2025) (NMI–RFE with Random Forest, 90.76%), the proposed STC-FSP achieves up to 16%–20% higher accuracy with significantly lower inference latency (~0.05s), confirming its superiority in both accuracy and computational efficiency. By fusing multi-domain descriptors and emphasizing low latency, the proposed approach is particularly suited to resource-constrained mental health monitoring applications. This study demonstrates the efficacy of interpretable handcrafted EEG features and log-correlation modelling for stress recognition.