https://reunir.unir.net/handle/123456789/17626 Wed, 14 May 2025 14:53:27 GMT 2025-05-14T14:53:27Z https://reunir.unir.net/handle/123456789/15030 An Improved Deep Learning Model for Electricity Price Forecasting Iqbal, Rashed; Mokhlis, Hazlie; Mohd Khairuddin, Anis Salwa; Azam Muhammad, Munir Accurate electricity price forecasting (EPF) is important for the purpose of bidding strategies and minimizing the risk for market participants in the competitive electricity market. Besides that, EPF becomes critically important for effective planning and efficient operation of a power system due to deregulation of electricity industry. However, accurate EPF is very challenging due to complex nonlinearity in the time series-based electricity prices. Hence, this work proposed two-fold contributions which are (1) effective time series preprocessing module to ensure feasible time-series data is fitted in the deep learning model, and (2) an improved long short-term memory (LSTM) model by incorporating linear scaled hyperbolic tangent (LiSHT) layer in the EPF. In this work, the time series pre-processing module adopted linear trend of the correlated features of electricity price series and the time series are tested by using Augmented Dickey Fuller (ADF) test method. In addition, the time series are transformed using boxcox transformation method in order to satisfy the stationarity property. Then, an improved LSTM prediction module is proposed to forecast electricity prices where LiSHT layer is adopted to optimize the parameters of the heterogeneous LSTM. This study is performed using the Australian electricity market price, load and renewable energy supply data. The experimental results obtained show that the proposed EPF framework performed better compared to previous techniques. Submitted by Susana Figueroa Navarro (susana.figueroa.n@unir.net) on 2023-07-11T10:59:08Z No. of bitstreams: 1 ip2023_06_001.pdf: 2496568 bytes, checksum: abb16105141372c6a491ec064dbf9a26 (MD5); Made available in DSpace on 2023-07-11T10:59:08Z (GMT). No. of bitstreams: 1 ip2023_06_001.pdf: 2496568 bytes, checksum: abb16105141372c6a491ec064dbf9a26 (MD5) https://reunir.unir.net/handle/123456789/15030 https://reunir.unir.net/handle/123456789/14813 Efficient Gated Convolutional Recurrent Neural Networks for Real-Time Speech Enhancement Fazal-E -Wahab; Ye, Zhongfu; Saleem, Nasir; Ali, Hamza Deep learning (DL) networks have grown into powerful alternatives for speech enhancement and have achieved excellent results by improving speech quality, intelligibility, and background noise suppression. Due to high computational load, most of the DL models for speech enhancement are difficult to implement for realtime processing. It is challenging to formulate resource efficient and compact networks. In order to address this problem, we propose a resource efficient convolutional recurrent network to learn the complex ratio mask for real-time speech enhancement. Convolutional encoder-decoder and gated recurrent units (GRUs) are integrated into the Convolutional recurrent network architecture, thereby formulating a causal system appropriate for real-time speech processing. Parallel GRU grouping and efficient skipped connection techniques are engaged to achieve a compact network. In the proposed network, the causal encoder-decoder is composed of five convolutional (Conv2D) and deconvolutional (Deconv2D) layers. Leaky linear rectified unit (ReLU) is applied to all layers apart from the output layer where softplus activation to confine the network output to positive is utilized. Furthermore, batch normalization is adopted after every convolution (or deconvolution) and prior to activation. In the proposed network, different noise types and speakers can be used in training and testing. With the LibriSpeech dataset, the experiments show that the proposed real-time approach leads to improved objective perceptual quality and intelligibility with much fewer trainable parameters than existing LSTM and GRU models. The proposed model obtained an average of 83.53% STOI scores and 2.52 PESQ scores, respectively. The quality and intelligibility are improved by 31.61% and 17.18% respectively over noisy speech. Submitted by Susana Figueroa Navarro (susana.figueroa.n@unir.net) on 2023-06-01T10:22:12Z No. of bitstreams: 1 ip2023_05_007.pdf: 2253358 bytes, checksum: 2f2676659a89cc17ef33fee27b97df02 (MD5); Made available in DSpace on 2023-06-01T10:22:12Z (GMT). No. of bitstreams: 1 ip2023_05_007.pdf: 2253358 bytes, checksum: 2f2676659a89cc17ef33fee27b97df02 (MD5) https://reunir.unir.net/handle/123456789/14813 https://reunir.unir.net/handle/123456789/14811 A Benchmark for the UEQ+ Framework: Construction of a Simple Tool to Quickly Interpret UEQ+ KPIs Meiners, Anna-Lena; Schrepp, Martin; Hinderks, Andreas; Thomaschewski, Jörg Questionnaires are a highly efficient method to compare the user experience (UX) of different interactive products or versions of a single product. Concretely, they allow us to evaluate the UX easily and to compare different products with a numeric UX score. However, often only one UX score from a single evaluated product is available. Without a comparison to other measurements, it is difficult to interpret an individual score, e.g. to decide whether a product’s UX is good enough to compete in the market. Many questionnaires offer benchmarks to support researchers in these cases. A benchmark is the result of a larger set of product evaluations performed with the same questionnaire. The score obtained from a single product evaluation can be compared to the scores from this benchmark data set to quickly interpret the results. In this paper, the first benchmark for the UEQ+ (User Experience Questionnaire +) is presented, which was created using 3.290 UEQ+ responses for 26 successful software products. The UEQ+ is a modular framework that contains a high number of validated user experience scales that can be combined to form a UX questionnaire. Currently, no benchmark is available for this framework, making the benchmark constructed in this paper a valuable interpretation tool for UEQ+ questionnaires. Submitted by Susana Figueroa Navarro (susana.figueroa.n@unir.net) on 2023-06-01T10:03:31Z No. of bitstreams: 1 ip2023_05_003.pdf: 956451 bytes, checksum: 1ebeb9b7b4361cfbbd2175e1e172f038 (MD5); Made available in DSpace on 2023-06-01T10:03:31Z (GMT). No. of bitstreams: 1 ip2023_05_003.pdf: 956451 bytes, checksum: 1ebeb9b7b4361cfbbd2175e1e172f038 (MD5) https://reunir.unir.net/handle/123456789/14811 https://reunir.unir.net/handle/123456789/14589 Analysis of Gender Differences in Facial Expression Recognition Based on Deep Learning Using Explainable Artificial Intelligence Manresa-Yee, Cristina; Ramis, Silvia; Buades, José M. Potential uses of automated Facial Expression Recognition (FER) cover a wide range of applications such as customer behavior analysis, healthcare applications or providing personalized services. Data for machine learning play a fundamental role, therefore, understanding the relevancy of the data in the outcomes is of utmost importance. In this work we present a study on how gender influences the learning of a FER system. We analyze with Explainable Artificial intelligence (XAI) techniques how gender contributes to the learning and assess which facial expressions are more similar regarding face regions that impact on the classification. Results show that there exist common regions in some expressions both for females and males with different intensities (e.g. happiness); however, there are other expressions like disgust, where important face regions differ. The insights of this work will help improving FER systems and understand the source of any inequality. Submitted by Susana Figueroa Navarro (susana.figueroa.n@unir.net) on 2023-05-03T09:22:35Z No. of bitstreams: 1 ip2023_04_003.pdf: 2400476 bytes, checksum: 29a450353bf34f3a9da5ddb4e01f8a22 (MD5); Made available in DSpace on 2023-05-03T09:22:35Z (GMT). No. of bitstreams: 1 ip2023_04_003.pdf: 2400476 bytes, checksum: 29a450353bf34f3a9da5ddb4e01f8a22 (MD5) https://reunir.unir.net/handle/123456789/14589 https://reunir.unir.net/handle/123456789/14313 A Feature Selection Approach Based on Archimedes’ Optimization Algorithm for Optimal Data Classification Khrissi, Lahbib; El Akkad, Nabil; Satori, Hassan; Satori, Khalid Feature selection is an active research area in data mining and machine learning, especially with the increase in the amount of numerical data. FS is a search strategy to find the best subset of features among a large number of subsets of features. Thus, FS is applied in most modern applications and in various domains, which requires the search for a powerful FS technique to process and classify high-dimensional data. In this paper, we propose a new technique for dimension reduction in feature selection. This approach is based on a recent metaheuristic called Archimedes’ Optimization Algorithm (AOA) to select an optimal subset of features to improve the classification accuracy. The idea of the AOA is based on the steps of Archimedes' principle in physics. It explains the behavior of the force exerted when an object is partially or fully immersed in a fluid. AOA optimization maintains a balance between exploration and exploitation, keeping a population of solutions and studying a large area to find the best overall solution. In this study, AOA is exploited as a search technique to find an optimal feature subset that reduces the number of features to maximize classification accuracy. The K-nearest neighbor (K-NN) classifier was used to evaluate the classification performance of selected feature subsets. To demonstrate the superiority of the proposed method, 16 benchmark datasets from the UCI repository are used and also compared by well-known and recently introduced meta-heuristics in this context, such as: sine-cosine algorithm (SCA), whale optimization algorithm (WOA), butterfly optimization algorithm (BAO), and butterfly flame optimization algorithm (MFO). The results prove the effectiveness of the proposed algorithm over the other algorithms based on several performance measures used in this paper. Submitted by Susana Figueroa Navarro (susana.figueroa.n@unir.net) on 2023-03-09T15:45:11Z No. of bitstreams: 1 ip2023_01_005.pdf: 1715840 bytes, checksum: db554ba969c6df88b4c141b7b1499932 (MD5); Made available in DSpace on 2023-03-09T15:45:11Z (GMT). No. of bitstreams: 1 ip2023_01_005.pdf: 1715840 bytes, checksum: db554ba969c6df88b4c141b7b1499932 (MD5) https://reunir.unir.net/handle/123456789/14313