Text-to-sound systems have recently attracted interest for their ability to synthesize common sounds from textual descriptions. However, previous research on sound generation has shown limited generation quality and increased computational complexity. We present SoundCrafter, a text-to-sound generation framework that utilizes diffusion models. Unlike previous methods, SoundCrafter operates within a compressed domain of mel spectrograms and is driven by semantic embeddings derived from the CLAP model, which stands for contrastive language audio pretraining. SoundCrafter improves generation quality and computational efficiency by learning the sound signals without modeling the cross-modal interaction. In addition, we employ a curricular learning technique by progressively increasing spectrogram resolution to stabilize training and improve output fidelity. SoundCrafter distinguishes itself by integrating CLAP-conditional semantic embeddings with a diffusion model that operates in the compressed domain of mel-spectrograms. Using the AudioCaps dataset, it achieves superior text-to-sound synthesis with a Fréchet Distance (FD) of 23.45 and an Inception Score (IS) of 7.57 - exceeding the performance of previous models while requiring significantly less computational resources and training on a single GPU.

Download article

• APA 7th style

​Alhaji, H., & Yaseen Taqa, A. (2025). SoundCrafter: Bridging text and Sound with a diffusion model. Applied Computer Science, 21(4), 13–20. https://doi.org/10.35784/acs_7549

• Chicago style

​Alhaji, Haitham, and Alaa Yaseen Taqa. ‘SoundCrafter: Bridging Text and Sound with a Diffusion Model’. Applied Computer Science 21, no. 4 (2025): 13–20. https://doi.org/10.35784/acs_7549.

• IEEE style

​H. Alhaji and A. Yaseen Taqa, ‘SoundCrafter: Bridging text and Sound with a diffusion model’, Applied Computer Science, vol. 21, no. 4, pp. 13–20, doi: 10.35784/acs_7549.

• Vancouver style

​Alhaji H, Yaseen Taqa A. SoundCrafter: Bridging text and Sound with a diffusion model. Applied Computer Science. 2025; 21(4):13–20.

Osteoarthritis (OA) is the most common joint disease and a leading cause of disability, most commonly affecting the knee. Conventional diagnostics rely primarily on imaging, which often detects changes only in advanced stages. This pilot study explores an alternative approach - encoder-based motion analysis combined with machine learning - to support early functional assessment of knee OA. The study included 90 subjects: 45 patients with radiographic evidence of OA and 45 healthy controls. A high-resolution rotary encoder integrated into a stabilizing knee orthosis recorded joint flexion-extension angles and velocities during open kinetic chain (OKC) and closed kinetic chain (CKC) tasks. Each subject performed five repetitions for each condition. Statistical analyses (Mann-Whitney U-test) revealed significant differences between groups, particularly in the CKC condition, where OA patients consistently required more time to complete movements. Machine learning classifiers were trained on cycle duration features. For OKC, accuracy remained modest (Naive Bayes: 65.6%), whereas CKC-based features provided stronger discrimination, with a narrow neural network achieving 80% accuracy and balanced sensitivity/specificity. The results demonstrate the feasibility of wearable encoder-based systems for objective, non-invasive assessment of knee function. CKC tasks showed higher diagnostic value, highlighting their potential for integration into clinical protocols. Future research should expand data sets, incorporate multimodal sensors, and use advanced algorithms to improve diagnostic performance and support real-world monitoring.

Download article

• APA 7th style

​Karpiński, R., & Syta, A. (2025). Application of encoder-based motion analysis and machine learning for knee osteoarthritis detection: A pilot study. Applied Computer Science, 21(4), 21–31. https://doi.org/10.35784/acs_8410

• Chicago style

​Karpiński, Robert, and Arkadiusz Syta. ‘Application of Encoder-Based Motion Analysis and Machine Learning for Knee Osteoarthritis Detection: A Pilot Study’. Applied Computer Science 21, no. 4 (2025): 21–31. https://doi.org/10.35784/acs_8410.

• IEEE style

​R. Karpiński and A. Syta, ‘Application of encoder-based motion analysis and machine learning for knee osteoarthritis detection: A pilot study’, Applied Computer Science, vol. 21, no. 4, pp. 21–31, doi: 10.35784/acs_8410.

• Vancouver style

​Karpiński R, Syta A. Application of encoder-based motion analysis and machine learning for knee osteoarthritis detection: A pilot study. Applied Computer Science. 2025; 21(4):21–31.

This study introduces SSAtt-SolNet, a novel deep learning approach designed to detect dusty solar panels, thereby improving the efficiency and reliability of solar photovoltaic systems. The proposed model uses MobileNetV3 as its backbone to balance accuracy and computational efficiency. We introduced a novel sparse shuffle block that combines depth-separable convolution with a shuffle layer to improve model performance. We also incorporated an attention mechanism in the classification layer to selectively focus on relevant features while minimizing noise interference. This lightweight approach was evaluated on two public and one self-collected dataset containing a total of 10,118 images. The model was benchmarked against eight SOTA models in image classification and dusty solar panel detection using four metrics: accuracy, model parameters, model size, and floating point operations (FLOPs). The experimental results showed that our approach outperformed all baseline models, achieving the smallest standard deviation over five folds (99.68 ± 0.3%). Furthermore, the proposed model had the smallest size, the fewest parameters, and the minimum GFLOPs (0.1005). The paired t-test confirmed that the accuracy of our model is statistically significantly higher than all baseline models at the 95% confidence level. These results suggest that our proposed model is feasible for use in environments with limited computing resources.

Download article

• APA 7th style

​Cong Tran, A., & Cong Tran, N. (2025). SSAtt-SolNet: An efficient model for dusty solar panel classification with Sparse Shuffle and Attention mechanisms. Applied Computer Science, 21(4), 32–46. https://doi.org/10.35784/acs_7589

• Chicago style

​Cong Tran, An, and Nghi Cong Tran. ‘SSAtt-SolNet: An Efficient Model for Dusty Solar Panel Classification with Sparse Shuffle and Attention Mechanisms’. Applied Computer Science 21, no. 4 (2025): 32–46. https://doi.org/10.35784/acs_7589.

• IEEE style

​A. Cong Tran and N. Cong Tran, ‘SSAtt-SolNet: An efficient model for dusty solar panel classification with Sparse Shuffle and Attention mechanisms’, Applied Computer Science, vol. 21, no. 4, pp. 32–46, doi: 10.35784/acs_7589.

• Vancouver style

​Cong Tran A, Cong Tran N. SSAtt-SolNet: An efficient model for dusty solar panel classification with Sparse Shuffle and Attention mechanisms. Applied Computer Science. 2025; 21(4):32–46.

An IoT sensor network enables real-time monitoring of key environmental parameters, including temperature, humidity, nutrient solution pH, and concentration. This system employs a rule-based expert system with dynamic threshold adjustments for automated control. Over a 35-day growth cycle (Days After Transplanting - DAT), the experiment revealed statistically significant improvement in lettuce growth within the smart indoor farming system compared to outdoor farming. Average increases were observed: 17.1% in plant weight, 16.9% in plant height, and 20.0% in the number of leaves. The IoT-based control system robustly maintained environmental parameters within optimal ranges, creating a stable and conducive growth environment. This approach highlights the transformative potential of integrating IoT and intelligent control logic for optimizing indoor hydroponic crop production, paving the way for more efficient and sustainable agriculture. The findings offer insights for future smart farming developments by demonstrating how IoT and intelligent control improve hydroponic lettuce yield.

Download article

• APA 7th style

​Barus, O. P., Maulana, A., Yugopuspito, P., Hidayanto, A. N., & Alexander, W. J. (2025). IoT-driven environmental optimization for hydroponic lettuce: A data-centric approach to smart agriculture. Applied Computer Science, 21(4), 47–58. https://doi.org/10.35784/acs_7793

• Chicago style

​Barus, Okky Putra, Ade Maulana, Pujianto Yugopuspito, Achmad Nizar Hidayanto, and Winar Joko Alexander. ‘IoT-Driven Environmental Optimization for Hydroponic Lettuce: A Data-Centric Approach to Smart Agriculture’. Applied Computer Science 21, no. 4 (2025): 47–58. https://doi.org/10.35784/acs_7793.

• IEEE style

​O. P. Barus, A. Maulana, P. Yugopuspito, A. N. Hidayanto, and W. J. Alexander, ‘IoT-driven environmental optimization for hydroponic lettuce: A data-centric approach to smart agriculture’, Applied Computer Science, vol. 21, no. 4, pp. 47–58, doi: 10.35784/acs_7793.

• Vancouver style

​Barus OP, Maulana A, Yugopuspito P, Hidayanto AN, Alexander WJ. IoT-driven environmental optimization for hydroponic lettuce: A data-centric approach to smart agriculture. Applied Computer Science. 2025; 21(4):47–58.

The paper presents an original method for processing medical data from a type 1 diabetes patient, with the aim of generating therapeutic recommendations to improve the quality of patient care. The article summarizes the results of the first phase of research in this area, which focused on identifying mathematical models and selecting algorithmic methods for further verification in clinical settings. The problem under study is characterized by high complexity, the need to tailor the method to the available data, and, in the completed stage of the research, the inability to perform experiments beyond computer simulations. The proposed approach introduces several novel solutions, including the development of a computer model of a person with diabetes, an original time-series similarity criterion for blood glucose concentration, and the innovative application of a genetic algorithm. The use of the genetic algorithm proved to be effective. The method was developed for patients using an insulin pump and a continuous glucose monitoring system. In the research section, data from five real patients were analyzed using the developed method, and the results indicated that it may be effective in supporting real-world therapy.

Download article

• APA 7th style

​Nowicki, T. (2025). Computer-Aided System with Machine Learning components for generating medical recommendations for type 1 diabetes patients. Applied Computer Science, 21(4), 59–75. https://doi.org/10.35784/acs_7950

• Chicago style

​Nowicki, Tomasz. ‘Computer-Aided System with Machine Learning Components for Generating Medical Recommendations for Type 1 Diabetes Patients’. Applied Computer Science 21, no. 4 (2025): 59–75. https://doi.org/10.35784/acs_7950.

• IEEE style

​T. Nowicki, ‘Computer-Aided System with Machine Learning components for generating medical recommendations for type 1 diabetes patients’, Applied Computer Science, vol. 21, no. 4, pp. 59–75, doi: 10.35784/acs_7950.

• Vancouver style

​Nowicki T. Computer-Aided System with Machine Learning components for generating medical recommendations for type 1 diabetes patients. Applied Computer Science. 2025; 21(4):59–75.

Rapid industrial growth in developing countries requires robust maintenance, and predictive maintenance (PdM) is a key solution to minimize downtime and costs. However, complex industrial systems and the acute scarcity of tagged data, particularly in African contexts, pose significant implementation challenges for traditional PdM approaches. This research proposes a novel predictive maintenance approach using a Variational Autoencoder (VAE) specifically designed to address data scarcity and improve interpretability in complex industrial systems in developing countries. The VAE is trained on real operational data and learns complex system patterns. Its interpretability is a key feature, achieved through visualization and analysis of latent space, providing deeper insight into system behavior. The VAE model demonstrates strong and consistent performance in anomaly detection and data reconstruction, as evidenced by low Mean Squared Error (MSE) and favorable R² values, and is rigorously validated through cross-validation, confirming its robustness and generalizability. This underscores its ability to accurately model complex system dynamics across diverse data subsets. This interpretable VAE model offers a powerful and promising predictive maintenance solution for improving the reliability of complex industrial systems in developing countries. By enabling early anomaly detection, synthetic data generation, and improved decision making, this approach has the potential to significantly contribute to the growth and sustainability of industries in these regions through reduced downtime and optimized resource utilization.

Download article

• APA 7th style

​Nasso Toumba, R., Moamissoal Samuel, M., Eboke, A., Taiwe, W., & Kombe, T. (2025). Interpretable VAE-based predictive modeling for enhanced complex industrial systems dependability in developing countries. Applied Computer Science, 21(4), 76–97. https://doi.org/10.35784/acs_7437

• Chicago style

​Nasso Toumba, Richard, Maxime Moamissoal Samuel, Achille Eboke, Wangkaké Taiwe, and Timothée Kombe. ‘Interpretable VAE-Based Predictive Modeling for Enhanced Complex Industrial Systems Dependability in Developing Countries’. Applied Computer Science 21, no. 4 (2025): 76–97. https://doi.org/10.35784/acs_7437.

• IEEE style

​R. Nasso Toumba, M. Moamissoal Samuel, A. Eboke, W. Taiwe, and T. Kombe, ‘Interpretable VAE-based predictive modeling for enhanced complex industrial systems dependability in developing countries’, Applied Computer Science, vol. 21, no. 4, pp. 76–97, doi: 10.35784/acs_7437.

• Vancouver style

​Nasso Toumba R, Moamissoal Samuel M, Eboke A, Taiwe W, Kombe T. Interpretable VAE-based predictive modeling for enhanced complex industrial systems dependability in developing countries. Applied Computer Science. 2025; 21(4):76–97.

In recent years, Data Envelopment Analysis (DEA) has gained popularity as a robust approach for assessing the eco-efficiency of economic units of different scales. This paper demonstrates the capabilities of the latest standalone version of the open-access MaxDEA X 12.2 software to measure comparative eco-efficiency, using the countries of the Eurasian Economic Union (EAEU) as a case study for the period 2015-2023. The study uses a traditional "black box" DEA model with atmospheric emissions, waste generation, and water consumption as inputs, and GDP along with population as outputs, allowing for a structural eco-efficiency assessment focused on resource use and economic structure. Calculation results obtained using the window method show that Belarus and Kyrgyzstan have the highest eco-efficiency over the entire observation window, while Kazakhstan and Russia lag behind, correlating with their natural resource-dependent economies. The analysis also provides target reductions in emissions and resource use for inefficient countries to improve eco-efficiency. In addition, the paper highlights how the MaxDEA X 12.2 software simplifies data handling and model configuration for eco-efficiency assessments by supporting different model orientations and returns to scale assumptions. Finally, it discusses potential extensions to more complex two-stage DEA models for comprehensive eco-efficiency assessments, subject to data availability. This work highlights the usefulness of MaxDEA X 12.2 as an accessible tool for eco-efficiency benchmarking and managerial decision support in the context of regional economic integration.

Download article

• APA 7th style

​Gabrielyan, B., Kesoyan, N., Ghazaryan, A., & Artashyan, A. (2025). Measuring comparative eco-efficiency in the Eurasian Economic Union using MaxDEA X 12.2 software. Applied Computer Science, 21(4), 98–109. https://doi.org/10.35784/acs_7702

• Chicago style

​Gabrielyan, Bella, Narek Kesoyan, Armen Ghazaryan, and Argam Artashyan. ‘Measuring Comparative Eco-Efficiency in the Eurasian Economic Union Using MaxDEA X 12.2 Software’. Applied Computer Science 21, no. 4 (2025): 98–109. https://doi.org/10.35784/acs_7702.

• IEEE style

​B. Gabrielyan, N. Kesoyan, A. Ghazaryan, and A. Artashyan, ‘Measuring comparative eco-efficiency in the Eurasian Economic Union using MaxDEA X 12.2 software’, Applied Computer Science, vol. 21, no. 4, pp. 98–109, doi: 10.35784/acs_7702.

• Vancouver style

​Gabrielyan B, Kesoyan N, Ghazaryan A, Artashyan A. Measuring comparative eco-efficiency in the Eurasian Economic Union using MaxDEA X 12.2 software. Applied Computer Science. 2025; 21(4):98–109.

Speech emotion recognition has been gaining importance for years, but most of the existing models are based on a single signal representation or conventional convolutional layers with a large number of parameters. In this study, we propose a compact multi-representation architecture that combines four images of the speech signal: spectrogram, MFCC features, wavelet scalogram, and fuzzy transform maps. Furthermore, the application of Kronecker convolution for efficient feature extraction with an extended receptive field is shown. Another novelty is cross-fusion, a mechanism that models interactions between branches without significantly increasing complexity. The core of the network is complemented by a transformer-based block and language-independent adversarial learning. The model is evaluated in a scenario of quadruple cross-lingual tests covering four data corpora for four languages: English, German, Polish and Danish. It is trained on three languages and tested on the fourth, achieving a weighted accuracy of 96.3%. In addition, the influence of selected activation functions on the classification quality is investigated. Ablation analysis shows that removing the Kronecker convolution reduces the efficiency by 5.6%, and removing the fuzzy transform representation by 4.7%. The obtained results indicate that the combination of Kronecker convolution, multi-channel fusion, and adversarial learning is a promising direction for building universal, language-independent emotion recognition systems.

Download article

• APA 7th style

​Powroźnik, P., & Skublewska-Paszkowska, M. (2025). K4F-Net: Lightweight multi-view speech emotion recognition with Kronecker convolution and cross-language robustness. Applied Computer Science, 21(4), 110–126. https://doi.org/10.35784/acs_8130

• Chicago style

​Powroźnik, Paweł, and Maria Skublewska-Paszkowska. ‘K4F-Net: Lightweight Multi-View Speech Emotion Recognition with Kronecker Convolution and Cross-Language Robustness’. Applied Computer Science 21, no. 4 (2025): 110–126. https://doi.org/10.35784/acs_8130.

• IEEE style

​P. Powroźnik and M. Skublewska-Paszkowska, ‘K4F-Net: Lightweight multi-view speech emotion recognition with Kronecker convolution and cross-language robustness’, Applied Computer Science, vol. 21, no. 4, pp. 110–126, doi: 10.35784/acs_8130.

• Vancouver style

​Powroźnik P, Skublewska-Paszkowska M. K4F-Net: Lightweight multi-view speech emotion recognition with Kronecker convolution and cross-language robustness. Applied Computer Science. 2025; 21(4):110–126.

Shape memory alloys (SMAs) exhibit several unique properties, including superelasticity and the shape memory effect. They can return to their original shape after deformation when heated. SMAs are widely used in various fields of science and technology. Shape memory alloys are functional materials that are used under loading, which in many cases is cyclic in nature. In the present study, the functional properties of NiTi shape memory alloys were modeled using supervised learning methods. The analysis was performed using Orange data mining software, which allows the creation of visual flowcharts and the generation of results in tables and graphs. The modeling was performed on four specimens. For each specimen, several functional properties, such as residual strain range ∆ε_r and dissipated energy range ∆W_dis. Each data set was divided into two unequal parts - the training and test sets. The training sets comprised 66% of the total data set. The remaining 34% was used for the test set. Among the methods studied, kNN, AdaBoost, Gradient Boosting and Random Forest showed the best results in terms of prediction errors. Therefore, ML learning methods are a powerful and promising tool for solving tasks related to the prediction of functional properties of SMAs.

Download article

• APA 7th style

​Hutsaylyuk, V., Demchyk, V., Yasniy, O., Lutsyk, N., & Fiialka, A. (2025). The modelling of NiTi shape memory alloy functional properties by machine learning methods. Applied Computer Science, 21(4), 127–135. https://doi.org/10.35784/acs_7986

• Chicago style

​Hutsaylyuk, Volodymyr, Vladyslav Demchyk, Oleh Yasniy, Nadiia Lutsyk, and Andrii Fiialka. ‘The Modelling of NiTi Shape Memory Alloy Functional Properties by Machine Learning Methods’. Applied Computer Science 21, no. 4 (2025): 127–135. https://doi.org/10.35784/acs_7986.

• IEEE style

​V. Hutsaylyuk, V. Demchyk, O. Yasniy, N. Lutsyk, and A. Fiialka, ‘The modelling of NiTi shape memory alloy functional properties by machine learning methods’, Applied Computer Science, vol. 21, no. 4, pp. 127–135, doi: 10.35784/acs_7986.

• Vancouver style

​Hutsaylyuk V, Demchyk V, Yasniy O, Lutsyk N, Fiialka A. The modelling of NiTi shape memory alloy functional properties by machine learning methods. Applied Computer Science. 2025; 21(4):127–135.

The study focuses on the development and evaluation of predictive models for forecasting labour demand in the metallurgical industry of the East Kazakhstan Region, with particular emphasis on the impact of production volume and labour productivity. The methodological framework combines classical econometric approaches with modern machine learning techniques, which makes it possible to capture nonlinear dependencies and more accurately assess labour market dynamics. The research is based on regional statistical data for the period 2015–2023. Several modeling approaches were tested, including linear regression, a parametric specification, and a hybrid machine learning model that integrates decision trees with local linear regression. Model performance was validated using the Mean Absolute Error (MAE), followed by forecasting labour demand for 2024–2028. Results demonstrate that the hybrid model outperforms the alternatives by achieving the lowest prediction error and producing the most plausible projection of moderate employment growth. The parametric model, although less precise, offers a high level of interpretability and is well suited for strategic analysis, while the linear regression model has limited effectiveness under nonlinear conditions. The practical value of the research lies in the possibility of embedding the developed models into decision support systems for government bodies and industrial enterprises, enabling early assessment of the impact of technological changes and production dynamics on employment. The outcomes may contribute to shaping balanced human resource policies, aligning educational programs with labour market needs, and conducting scenario analyses. Furthermore, the findings establish a foundation for extending the methodology to other industries and incorporating additional variables related to digitalization and innovation activity.

Download article

• APA 7th style

​Denissova, O., Ismukhamedov, A., Konurbayeva, Z., Rakhmetullina, S., Samussenko, Y., & Kulisz, M. (2025). Application of machine learning algorithms for forecasting labour demand in the metallurgical industry of the east Kazakhstan region. Applied Computer Science, 21(4), 136–158. https://doi.org/10.35784/8290

• Chicago style

​Denissova, Oxana, Aman Ismukhamedov, Zhadyra Konurbayeva, Saule Rakhmetullina, Yelena Samussenko, and Monika Kulisz. ‘Application of Machine Learning Algorithms for Forecasting Labour Demand in the Metallurgical Industry of the East Kazakhstan Region’. Applied Computer Science 21, no. 4 (2025): 136–158. https://doi.org/10.35784/8290.

• IEEE style

​O. Denissova, A. Ismukhamedov, Z. Konurbayeva, S. Rakhmetullina, Y. Samussenko, and M. Kulisz, ‘Application of machine learning algorithms for forecasting labour demand in the metallurgical industry of the east Kazakhstan region’, Applied Computer Science, vol. 21, no. 4, pp. 136–158, doi: 10.35784/8290.

• Vancouver style

​Denissova O, Ismukhamedov A, Konurbayeva Z, Rakhmetullina S, Samussenko Y, Kulisz M. Application of machine learning algorithms for forecasting labour demand in the metallurgical industry of the east Kazakhstan region. Applied Computer Science. 2025; 21(4):136–158.

The architectural design of deep learning models significantly influences their predictive capabilities in environmental monitoring tasks. This paper investigates the individual and collective effects of residual learning and feature fusion mechanism to improve the performance of soil moisture estimation on the designed architecture of the deep learning model. In this study, the data fusion mechanism was used to integrate Normalized Difference Water Index (NDWI), Synthetic Aperture Radar (SAR), and satellite imagery datasets containing Red, Green, and Blue (RGB) color channels, which consist of images or data collected by a radar system that uses microwaves to produce images of the Earth's surface. Three model variants were developed, each selectively omitting one or more of these architectural elements, and their performance was evaluated using three standard metrics, Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R2). The results of the final proposed model architecture showed that while each component contributes to accuracy improvements, the combination of residual learning and feature fusion yields the most significant gains. Improved results of RMSE = 0.0117, R²=0.814 and Mean Absolute Error =0.0148 were obtained. These performance indicators were superior to the results of most of the baseline models after comparative analysis. Thus, this study provides insights into model component selection for deep learning soil moisture prediction applications.

Download article

• APA 7th style

​Yamakili, P., Nicholaus, M. R., & Greyson, K. A. (2025). Evaluating the impact of residual learning and feature fusion on soil moisture prediction accuracy. Applied Computer Science, 21(4), 159–168. https://doi.org/10.35784/acs_7752

• Chicago style

​Yamakili, Pascal, Mrindoko Rashid Nicholaus, and Kenedy Aliila Greyson. ‘Evaluating the Impact of Residual Learning and Feature Fusion on Soil Moisture Prediction Accuracy’. Applied Computer Science 21, no. 4 (2025): 159–168. https://doi.org/10.35784/acs_7752.

• IEEE style

​P. Yamakili, M. R. Nicholaus, and K. A. Greyson, ‘Evaluating the impact of residual learning and feature fusion on soil moisture prediction accuracy’, Applied Computer Science, vol. 21, no. 4, pp. 159–168, doi: 10.35784/acs_7752.

• Vancouver style

​Yamakili P, Nicholaus MR, Greyson KA. Evaluating the impact of residual learning and feature fusion on soil moisture prediction accuracy. Applied Computer Science. 2025; 21(4):159–168.