Search results for: long short-term memory (lstm)

Authors: Heri Kuswanto, Philipp Sibbertsen, Irhamah

Abstract:

A Wald type test to distinguish between long memory and ESTAR nonlinearity has been developed. The test uses a directed-Wald statistic to overcome the problem of restricted parameters under the alternative. The test is derived from a model specification i.e. allows the transition parameter to appear as a nuisance parameter in the transition function. A simulation study has been conducted and it indicates that the approach leads a test with good size and power properties to distinguish between stationary long memory and ESTAR.

Keywords: directed-Wald test, ESTAR, long memory, distinguish

Procedia PDF Downloads 482

Authors: A. Boateng, La Gil-Alana, M. Lesaoana; Hj. Siweya, A. Belete

Abstract:

This study examines long memory or long-range dependence in the CPI inflation rates of Ghana and South Africa using Whittle methods and autoregressive fractionally integrated moving average (ARFIMA) models. Standard I(0)/I(1) methods such as Augmented Dickey-Fuller (ADF), Philips-Perron (PP) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) tests were also employed. Our findings indicate that long memory exists in the CPI inflation rates of both countries. After processing fractional differencing and determining the short memory components, the models were specified as ARFIMA (4,0.35,2) and ARFIMA (3,0.49,3) respectively for Ghana and South Africa. Consequently, the CPI inflation rates of both countries are fractionally integrated and mean reverting. The implication of this result will assist in policy formulation and identification of inflationary pressures in an economy.

Keywords: Consumer Price Index (CPI) inflation rates, Whittle method, long memory, ARFIMA model

Procedia PDF Downloads 370

Authors: Abdulaziz Afandi

Abstract:

Minimizing the inventory cost, optimizing the inventory quantities, and increasing system operational availability are the main motivations to enhance forecasting demand of spare parts in a major power utility company in Medina. This paper reports in an effort made to optimize the orders quantities of spare parts by improving the method of forecasting the demand. The study focuses on equipment that has frequent spare parts purchase orders with uncertain demand. The pattern of the demand considers a lumpy pattern which makes conventional forecasting methods less effective. A comparison was made by benchmarking various methods of forecasting based on experts’ criteria to select the most suitable method for the case study. Three actual data sets were used to make the forecast in this case study. Two neural networks (NN) approaches were utilized and compared, namely long short-term memory (LSTM) and multilayer perceptron (MLP). The results as expected, showed that the NN models gave better results than traditional forecasting method (judgmental method). In addition, the LSTM model had a higher predictive accuracy than the MLP model.

Keywords: neural network, LSTM, MLP, forecasting demand, inventory management

Procedia PDF Downloads 127

Authors: Zhe Li, Aihua Cai

Abstract:

Aiming at the problem of high-density clutter and interference affecting radar detection target track initiation in ECM and complex radar mission, the traditional radar target track initiation method has been difficult to adapt. To this end, we propose a multi-algorithm fusion learning track initiation algorithm, which transforms the track initiation problem into a true-false track discrimination problem, and designs an algorithm based on 1DCNN(One-Dimensional CNN)combined with Bi-LSTM (Bi-Directional Long Short-Term Memory )for fusion classification. The experimental dataset consists of real trajectories obtained from a certain type of three-coordinate radar measurements, and the experiments are compared with traditional trajectory initiation methods such as rule-based method, logical-based method and Hough-transform-based method. The simulation results show that the overall performance of the multi-algorithm fusion learning track initiation algorithm is significantly better than that of the traditional method, and the real track initiation rate can be effectively improved under high clutter density with the average initiation time similar to the logical method.

Keywords: track initiation, multi-algorithm fusion, 1DCNN, Bi-LSTM

Procedia PDF Downloads 95

Authors: Davide Fuscà, Kanan Rahimli, Roberto Leuzzi

Abstract:

Preventing abuse and illegal activities in a given area of the sea is a very difficult and expensive task. Artificial intelligence offers the possibility to implement new methods to identify the vessel class type from the kinematic features of the vessel itself. The task strictly depends on the quality of the data. This paper explores the application of a deep, long short-term memory model by using AIS flow only with a relatively low quality. The proposed model reaches high accuracy on detecting nine vessel classes representing the most common vessel types in the Ionian-Adriatic Sea. The model has been applied during the Adriatic-Ionian trial period of the international EU ANDROMEDA H2020 project to identify vessels performing behaviors far from the expected one depending on the declared type.

Keywords: maritime surveillance, artificial intelligence, behavior analysis, LSTM

Procedia PDF Downloads 231

Authors: Sima Najafzadehkhoei, George Vega Yon

Abstract:

This paper presents an approach for calibrating Agent-Based Models (ABMs) efficiently, utilizing Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks. These machine learning techniques are applied to Susceptible-Infected-Recovered (SIR) models, which are a core framework in the study of epidemiology. Our method replicates parameter values from observed trajectory curves, enhancing the accuracy of predictions when compared to traditional calibration techniques. Through the use of simulated data, we train the models to predict epidemiological parameters more accurately. Two primary approaches were explored: one where the number of susceptible, infected, and recovered individuals is fully known, and another using only the number of infected individuals. Our method shows promise for application in other ABMs where calibration is computationally intensive and expensive.

Keywords: ABM, calibration, CNN, LSTM, epidemiology

Procedia PDF Downloads 25

Authors: Amartya Hatua, Trung Nguyen, Andrew Sung

Abstract:

In this paper, we study the information diffusion process on Twitter as a multivariate time series problem. Our model concerns three measures (volume, network influence, and sentiment of tweets) based on 10 features, and we collected 27 million tweets to build our information diffusion time series dataset for analysis. Then, different time series clustering techniques with Dynamic Time Warping (DTW) distance were used to identify different patterns of information diffusion. Finally, we built the information diffusion prediction models for new hashtags which comprise two phrases: The first phrase is recognizing the pattern using k-NN with DTW distance; the second phrase is building the forecasting model using the traditional Autoregressive Integrated Moving Average (ARIMA) model and the non-linear recurrent neural network of Long Short-Term Memory (LSTM). Preliminary results of performance evaluation between different forecasting models show that LSTM with clustering information notably outperforms other models. Therefore, our approach can be applied in real-world applications to analyze and predict the information diffusion characteristics of selected topics or memes (hashtags) in Twitter.

Keywords: ARIMA, DTW, information diffusion, LSTM, RNN, time series clustering, time series forecasting, Twitter

Procedia PDF Downloads 391

Authors: Amir Andria Gad Shehata

Abstract:

Minimizing the inventory cost, optimizing the inventory quantities, and increasing system operational availability are the main motivations to enhance forecasting demand of spare parts in a major power utility company in Medina. This paper reports in an effort made to optimize the orders quantities of spare parts by improving the method of forecasting the demand. The study focuses on equipment that has frequent spare parts purchase orders with uncertain demand. The pattern of the demand considers a lumpy pattern which makes conventional forecasting methods less effective. A comparison was made by benchmarking various methods of forecasting based on experts’ criteria to select the most suitable method for the case study. Three actual data sets were used to make the forecast in this case study. Two neural networks (NN) approaches were utilized and compared, namely long short-term memory (LSTM) and multilayer perceptron (MLP). The results as expected, showed that the NN models gave better results than traditional forecasting method (judgmental method). In addition, the LSTM model had a higher predictive accuracy than the MLP model.

Keywords: spare part, spare part inventory, inventory model, optimization, maintenanceneural network, LSTM, MLP, forecasting demand, inventory management

Procedia PDF Downloads 63

Authors: Sonia Perez-Gamboa, Qingquan Sun, Yan Zhang

Abstract:

Non-intrusive sensor-based human activity recognition (HAR) is utilized in a spectrum of applications, including fitness tracking devices, gaming, health care monitoring, and smartphone applications. Deep learning models such as convolutional neural networks (CNNs) and long short term memory (LSTM) recurrent neural networks (RNNs) provide a way to achieve HAR accurately and effectively. In this paper, we design a multi-layer hybrid architecture with CNN and LSTM and explore a variety of multi-layer combinations. Based on the exploration, we present a lightweight, hybrid, and multi-layer model, which can improve the recognition performance by integrating local features and scale-invariant with dependencies of activities. The experimental results demonstrate the efficacy of the proposed model, which can achieve a 94.7% activity recognition rate on a benchmark human activity dataset. This model outperforms traditional machine learning and other deep learning methods. Additionally, our implementation achieves a balance between recognition rate and training time consumption.

Keywords: deep learning, LSTM, CNN, human activity recognition, inertial sensor

Procedia PDF Downloads 150

Authors: Elham Koohikerade, Silvio Jose Gumiere

Abstract:

In this research, irrigation is highlighted as crucial for improving both the yield and quality of ‎potatoes due to their high sensitivity to soil moisture changes. The study presents a hybrid Long ‎Short-Term Memory (LSTM) model aimed at optimizing irrigation scheduling in potato fields in ‎Quebec City, Canada. This model integrates model-based and satellite-derived datasets to simulate ‎soil moisture content, addressing the limitations of field data. Developed under the guidance of the ‎Food and Agriculture Organization (FAO), the simulation approach compensates for the lack of direct ‎soil sensor data, enhancing the LSTM model's predictions. The model was calibrated using indices ‎like Surface Soil Moisture (SSM), Normalized Vegetation Difference Index (NDVI), Enhanced ‎Vegetation Index (EVI), and Normalized Multi-band Drought Index (NMDI) to effectively forecast ‎soil moisture reductions. Understanding soil moisture and plant development is crucial for assessing ‎drought conditions and determining irrigation needs. This study validated the spectral characteristics ‎of vegetation and soil using ECMWF Reanalysis v5 (ERA5) and Moderate Resolution Imaging ‎Spectrometer (MODIS) data from 2019 to 2023, collected from agricultural areas in Dolbeau and ‎Peribonka, Quebec. Parameters such as surface volumetric soil moisture (0-7 cm), NDVI, EVI, and ‎NMDI were extracted from these images. A regional four-year dataset of soil and vegetation moisture ‎was developed using a machine learning approach combining model-based and satellite-based ‎datasets. The LSTM model predicts soil moisture dynamics hourly across different locations and ‎times, with its accuracy verified through cross-validation and comparison with existing soil moisture ‎datasets. The model effectively captures temporal dynamics, making it valuable for applications ‎requiring soil moisture monitoring over time, such as anomaly detection and memory analysis. By ‎identifying typical peak soil moisture values and observing distribution shapes, irrigation can be ‎scheduled to maintain soil moisture within Volumetric Soil Moisture (VSM) values of 0.25 to 0.30 ‎m²/m², avoiding under and over-watering. The strong correlations between parcels suggest that a ‎uniform irrigation strategy might be effective across multiple parcels, with adjustments based on ‎specific parcel characteristics and historical data trends. The application of the LSTM model to ‎predict soil moisture and vegetation indices yielded mixed results. While the model effectively ‎captures the central tendency and temporal dynamics of soil moisture, it struggles with accurately ‎predicting EVI, NDVI, and NMDI.‎

Keywords: irrigation scheduling, LSTM neural network, remotely sensed indices, soil and vegetation ‎monitoring

Procedia PDF Downloads 42

Authors: B. Sabayan, A. Alidadi, S. Ebrahimi, N. M. Bakhshani

Abstract:

Neuropsychological problems are more common in hemodialysis (HD) patients than in healthy individuals. The aim of this study was to investigate the effect of long term HD on memory types of HD patients. To assess the different type of memory, we used memory parts of the Persian Papers and Pencil Cognitive assessment package (PCAP) and Addenbrooke's Cognitive Examination (ACE-R). Our study included 80 HD patients of whom 39 had less than six months of HD and 41 patients and another group which had a history of HD more than six months. The population had a mean age of 51.60 years old and 27.5% of them were female. The scores of patients who have been hemodialyzed for a long time (median time of HD was up to 4 years) had lower score in anterograde, explicit, visual, recall and recognition memory (5.44±1.07, 9.49±3.472, 22.805±6.6913, 5.59±10.435, 11.02±3.190 score) than the HD patients who underwent HD for a shorter term, where the median time was 3 to 5 months (P<0.01). The regression result shows that, by increasing the HD duration, all memory types are reduced (R2=0.600, P<0.01). The present study demonstrated that HD patients who were under HD for a long time had significantly lower scores in the different types of memory. However, additional researches are needed in this area.

Keywords: hemodialysis patients, duration of hemodialysis, memory types, Zahedan

Procedia PDF Downloads 178

Authors: Sam Khozama, Ali M. Mayya

Abstract:

Breast cancer is one of the most common types in women. Early prediction of breast cancer helps physicians detect cancer in its early stages. Big cancer data needs a very powerful tool to analyze and extract predictions. Machine learning and deep learning are two of the most efficient tools for predicting cancer based on textual data. In this study, we developed a fusion model of two machine learning and deep learning models. To obtain the final prediction, Long-Short Term Memory (LSTM) and ensemble learning with hyper parameters optimization are used, and score-level fusion is used. Experiments are done on the Breast Cancer Surveillance Consortium (BCSC) dataset after balancing and grouping the class categories. Five different training scenarios are used, and the tests show that the designed fusion model improved the performance by 3.3% compared to the individual models.

Keywords: machine learning, deep learning, cancer prediction, breast cancer, LSTM, fusion

Procedia PDF Downloads 163

Authors: Anita Sadat Sadati Rostami, Reza Almasi Ghaleh

Abstract:

Surface electromyography (sEMG) offers important insights into muscle activation and has applications in fields including rehabilitation and human-computer interaction. The purpose of this work is to predict the degree of activation of two joints in the index finger using an LSTM-Transformer architecture trained on sEMG data from the Ninapro DB8 dataset. We apply advanced preprocessing techniques, such as multi-band filtering and customizable rectification methods, to enhance the encoding of sEMG data into features that are beneficial for regression tasks. The processed data is converted into spike patterns and simulated using Leaky Integrate-and-Fire (LIF) neuron models, allowing for neuromorphic-inspired processing. Our findings demonstrate that adjusting filtering parameters and neuron dynamics and employing the LSTM-Transformer model improves joint angle prediction performance. This study contributes to the ongoing development of deep learning frameworks for sEMG analysis, which could lead to improvements in motor control systems.

Keywords: surface electromyography, LSTM-transformer, spiking neural networks, hand kinematics, leaky integrate-and-fire neuron, band-pass filtering, muscle activity decoding

Procedia PDF Downloads 11

Authors: Suriya

Abstract:

Rapid socio-economic development and urbanization have led to an increasingly serious air pollution problem in Ulaanbaatar (UB), the capital of Mongolia. PM₂.₅ pollution has become the most pressing aspect of UB air pollution. Therefore, monitoring and predicting PM₂.₅ concentration in UB is of great significance for the health of the local people and environmental management. As of yet, very few studies have used models to predict PM₂.₅ concentrations in UB. Using data from 0:00 on June 1, 2018, to 23:00 on April 30, 2020, we proposed two deep learning models based on Bayesian-optimized LSTM (Bayes-LSTM) and CNN-LSTM. We utilized hourly observed data, including Himawari8 (H8) aerosol optical depth (AOD), meteorology, and PM₂.₅ concentration, as input for the prediction of PM₂.₅ concentrations. The correlation strengths between meteorology, AOD, and PM₂.₅ were analyzed using the gray correlation analysis method; the comparison of the performance improvement of the model by using the AOD input value was tested, and the performance of these models was evaluated using mean absolute error (MAE) and root mean square error (RMSE). The prediction accuracies of Bayes-LSTM and CNN-LSTM deep learning models were both improved when AOD was included as an input parameter. Improvement of the prediction accuracy of the CNN-LSTM model was particularly enhanced in the non-heating season; in the heating season, the prediction accuracy of the Bayes-LSTM model slightly improved, while the prediction accuracy of the CNN-LSTM model slightly decreased. We propose two novel deep learning models for PM₂.₅ concentration prediction in UB, Bayes-LSTM, and CNN-LSTM deep learning models. Pioneering the use of AOD data from H8 and demonstrating the inclusion of AOD input data improves the performance of our two proposed deep learning models.

Keywords: deep learning, AOD, PM2.5, prediction, Ulaanbaatar

Procedia PDF Downloads 48

Authors: Akinbo Razak Yinka, Adesanya Kehinde Kazeem, Oladokun Oluwagbenga Peter

Abstract:

Time series data are sequences of observations collected over a period of time. Such data can be used to predict health outcomes, such as disease progression, mortality, hospitalization, etc. The Statistical approach is based on mathematical models that capture the patterns and trends of the data, such as autocorrelation, seasonality, and noise, while Neural methods are based on artificial neural networks, which are computational models that mimic the structure and function of biological neurons. This paper compared both parametric and non-parametric time series models of patients treated for malaria in Maternal and Child Health Centres in Lagos State, Nigeria. The forecast methods considered linear regression, Integrated Moving Average, ARIMA and SARIMA Modeling for the parametric approach, while Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) Network were used for the non-parametric model. The performance of each method is evaluated using the Mean Absolute Error (MAE), R-squared (R2) and Root Mean Square Error (RMSE) as criteria to determine the accuracy of each model. The study revealed that the best performance in terms of error was found in MLP, followed by the LSTM and ARIMA models. In addition, the Bootstrap Aggregating technique was used to make robust forecasts when there are uncertainties in the data.

Keywords: ARIMA, bootstrap aggregation, MLP, LSTM, SARIMA, time-series analysis

Procedia PDF Downloads 75

Authors: Serpil Türkyılmaz, Mesut Balıbey

Abstract:

In this study, long memory properties in volatility of Turkey Stock Market are being examined through the FIGARCH, FIEGARCH and FIAPARCH models under different distribution assumptions as normal and skewed student-t distributions. Furthermore, structural changes in volatility of Turkey Stock Market are investigated. The results display long memory property and the presence of asymmetric effects of shocks in volatility of Turkey Stock Market.

Keywords: FIAPARCH model, FIEGARCH model, FIGARCH model, structural break

Procedia PDF Downloads 291

Authors: Shibo Wei, Ting Jiang

Abstract:

Convolutional-recurrent neural networks (CRN) have achieved much success recently in the speech enhancement field. The common processing method is to use the convolution layer to compress the feature space by multiple upsampling and then model the compressed features with the LSTM layer. At last, the enhanced speech is obtained by deconvolution operation to integrate the global information of the speech sequence. However, the feature space compression process may cause the loss of information, so we propose to model the upsampling result of each step with the residual LSTM layer, then join it with the output of the deconvolution layer and input them to the next deconvolution layer, by this way, we want to integrate the global information of speech sequence better. The experimental results show the network model (RES-CRN) we introduce can achieve better performance than LSTM without residual and overlaying LSTM simply in the original CRN in terms of scale-invariant signal-to-distortion ratio (SI-SNR), speech quality (PESQ), and intelligibility (STOI).

Keywords: convolutional-recurrent neural networks, speech enhancement, residual LSTM, SI-SNR

Procedia PDF Downloads 201

Authors: Yingkun Huang, Li Guo, Zekang Lan, Kai Tian

Abstract:

Hard disk drives (HDD) failure of the exascale supercomputing system may lead to service interruption and invalidate previous calculations, and it will cause permanent data loss. Therefore, initiating corrective actions before hard drive failures materialize is critical to the continued operation of jobs. In this paper, a highly accurate analysis model based on CNN-LSTM and oversampling technique was proposed, which can correctly predict the necessity of a disk replacement even ten days in advance. Generally, the learning-based method performs poorly on a training dataset with long-tail distribution, especially fault prediction is a very classic situation as the scarcity of failure data. To overcome the puzzle, a new oversampling was employed to augment the data, and then, an improved CNN-LSTM with the shortcut was built to learn more effective features. The shortcut transmits the results of the previous layer of CNN and is used as the input of the LSTM model after weighted fusion with the output of the next layer. Finally, a detailed, empirical comparison of 6 prediction methods is presented and discussed on a public dataset for evaluation. The experiments indicate that the proposed method predicts disk failure with 0.91 Precision, 0.91 Recall, 0.91 F-measure, and 0.90 MCC for 10 days prediction horizon. Thus, the proposed algorithm is an efficient algorithm for predicting HDD failure in supercomputing.

Keywords: HDD replacement, failure, CNN-LSTM, oversampling, prediction

Procedia PDF Downloads 80

Authors: Jacqueline Rose T. Alipo-on, Francesca Isabelle F. Escobar, Myles Joshua T. Tan, Hezerul Abdul Karim, Nouar Al Dahoul

Abstract:

Electrocardiogram (ECG) signal analysis and processing are crucial in the diagnosis of cardiovascular diseases, which are considered one of the leading causes of mortality worldwide. However, the traditional rule-based analysis of large volumes of ECG data is time-consuming, labor-intensive, and prone to human errors. With the advancement of the programming paradigm, algorithms such as machine learning have been increasingly used to perform an analysis of ECG signals. In this paper, various deep learning algorithms were adapted to classify five classes of heartbeat types. The dataset used in this work is the synthetic MIT-BIH Arrhythmia dataset produced from generative adversarial networks (GANs). Various deep learning models such as ResNet-50 convolutional neural network (CNN), 1-D CNN, and long short-term memory (LSTM) were evaluated and compared. ResNet-50 was found to outperform other models in terms of recall and F1 score using a five-fold average score of 98.88% and 98.87%, respectively. 1-D CNN, on the other hand, was found to have the highest average precision of 98.93%.

Keywords: heartbeat classification, convolutional neural network, electrocardiogram signals, generative adversarial networks, long short-term memory, ResNet-50

Procedia PDF Downloads 128

Authors: Thaveesha Dheerasekera, Dileeka Sandamali Alwis

Abstract:

The escalating prevalence of mental health issues, such as depression and suicidal ideation, is a matter of significant global concern. It is plausible that a variety of factors, such as life events, social isolation, and preexisting physiological or psychological health conditions, could instigate or exacerbate these conditions. Traditional approaches to diagnosing depression entail a considerable amount of time and necessitate the involvement of adept practitioners. This underscores the necessity for automated systems capable of promptly detecting and diagnosing symptoms of depression. The PsyVBot system employs sophisticated natural language processing and machine learning methodologies, including the use of the NLTK toolkit for dataset preprocessing and the utilization of a Long Short-Term Memory (LSTM) model. The PsyVBot exhibits a remarkable ability to diagnose depression with a 94% accuracy rate through the analysis of user input. Consequently, this resource proves to be efficacious for individuals, particularly those enrolled in academic institutions, who may encounter challenges pertaining to their psychological well-being. The PsyVBot employs a Long Short-Term Memory (LSTM) model that comprises a total of three layers, namely an embedding layer, an LSTM layer, and a dense layer. The stratification of these layers facilitates a precise examination of linguistic patterns that are associated with the condition of depression. The PsyVBot has the capability to accurately assess an individual's level of depression through the identification of linguistic and contextual cues. The task is achieved via a rigorous training regimen, which is executed by utilizing a dataset comprising information sourced from the subreddit r/SuicideWatch. The diverse data present in the dataset ensures precise and delicate identification of symptoms linked with depression, thereby guaranteeing accuracy. PsyVBot not only possesses diagnostic capabilities but also enhances the user experience through the utilization of audio outputs. This feature enables users to engage in more captivating and interactive interactions. The PsyVBot platform offers individuals the opportunity to conveniently diagnose mental health challenges through a confidential and user-friendly interface. Regarding the advancement of PsyVBot, maintaining user confidentiality and upholding ethical principles are of paramount significance. It is imperative to note that diligent efforts are undertaken to adhere to ethical standards, thereby safeguarding the confidentiality of user information and ensuring its security. Moreover, the chatbot fosters a conducive atmosphere that is supportive and compassionate, thereby promoting psychological welfare. In brief, PsyVBot is an automated conversational agent that utilizes an LSTM model to assess the level of depression in accordance with the input provided by the user. The demonstrated accuracy rate of 94% serves as a promising indication of the potential efficacy of employing natural language processing and machine learning techniques in tackling challenges associated with mental health. The reliability of PsyVBot is further improved by the fact that it makes use of the Reddit dataset and incorporates Natural Language Toolkit (NLTK) for preprocessing. PsyVBot represents a pioneering and user-centric solution that furnishes an easily accessible and confidential medium for seeking assistance. The present platform is offered as a modality to tackle the pervasive issue of depression and the contemplation of suicide.

Keywords: chatbot, depression diagnosis, LSTM model, natural language process

Procedia PDF Downloads 69

Authors: Farzad Hosseini Hossein Abadi, Cristina Prieto Sierra, Cesar Álvarez Díaz

Abstract:

Predicting streamflow accurately in flashy catchments prone to floods is a major research and operational challenge in hydrological modeling. Recent advancements in deep learning, particularly Long Short-Term Memory (LSTM) networks, have shown to be promising in achieving accurate hydrological predictions at daily and hourly time scales. In this work, a multi-timescale LSTM (MTS-LSTM) network was applied to the context of regional hydrological predictions at an hourly time scale in flashy catchments. The case study includes 40 catchments allocated in the Basque Country, north of Spain. We explore the impact of hyperparameters on the performance of streamflow predictions given by regional deep learning models through systematic hyperparameter tuning - where optimal regional values for different catchments are identified. The results show that predictions are highly accurate, with Nash-Sutcliffe (NSE) and Kling-Gupta (KGE) metrics values as high as 0.98 and 0.97, respectively. A principal component analysis reveals that a hyperparameter related to the length of the input sequence contributes most significantly to the prediction performance. The findings suggest that input sequence lengths have a crucial impact on the model prediction performance. Moreover, employing catchment-scale analysis reveals distinct sequence lengths for individual basins, highlighting the necessity of customizing this hyperparameter based on each catchment’s characteristics. This aligns with well known “uniqueness of the place” paradigm. In prior research, tuning the length of the input sequence of LSTMs has received limited focus in the field of streamflow prediction. Initially it was set to 365 days to capture a full annual water cycle. Later, performing limited systematic hyper-tuning using grid search, revealed a modification to 270 days. However, despite the significance of this hyperparameter in hydrological predictions, usually studies have overlooked its tuning and fixed it to 365 days. This study, employing a simultaneous systematic hyperparameter tuning approach, emphasizes the critical role of input sequence length as an influential hyperparameter in configuring LSTMs for regional streamflow prediction. Proper tuning of this hyperparameter is essential for achieving accurate hourly predictions using deep learning models.

Keywords: LSTMs, streamflow, hyperparameters, hydrology

Procedia PDF Downloads 70

Authors: Liang Wang, Beste F. Yuksel, David Guy Brizan

Abstract:

A computational model of affect that can distinguish between spontaneous and posed smiles with no errors on a large, popular data set using deep learning techniques is presented in this paper. A Long Short-Term Memory (LSTM) classifier, a type of Recurrent Neural Network, is utilized and compared to human classification. Results showed that while human classification (mean of 0.7133) was above chance, the LSTM model was more accurate than human classification and other comparable state-of-the-art systems. Additionally, a high accuracy rate was maintained with small amounts of training videos (70 instances). The derivation of important features to further understand the success of our computational model were analyzed, and it was inferred that thousands of pairs of points within the eyes and mouth are important throughout all time segments in a smile. This suggests that distinguishing between a posed and spontaneous smile is a complex task, one which may account for the difficulty and lower accuracy of human classification compared to machine learning models.

Keywords: affective computing, affect detection, computer vision, deep learning, human-computer interaction, machine learning, posed smile detection, spontaneous smile detection

Procedia PDF Downloads 125

Authors: Rafael Marinho de Andrade, Elcio Hideti Shiguemori, Rafael Duarte Coelho dos Santos

Abstract:

Relying on computer vision, many clever things are possible in order to make the world safer and optimized on resource management, especially considering time and attention as manageable resources, once the modern world is very abundant in cameras from inside our pockets to above our heads while crossing the streets. Thus, automated solutions based on computer vision techniques to detect, react, or even prevent relevant events such as robbery, car crashes and traffic jams can be accomplished and implemented for the sake of both logistical and surveillance improvements. In this paper, we present an approach for vehicles’ abnormal behaviors detection from highway footages, in which the vectorial data of the vehicles’ displacement are extracted directly from surveillance cameras footage through object detection and tracking with a deep convolutional neural network and inserted into a long-short term memory neural network for behavior classification. The results show that the classifications of behaviors are consistent and the same principles may be applied to other trackable objects and scenarios as well.

Keywords: artificial intelligence, behavior detection, computer vision, convolutional neural networks, LSTM, highway footage

Procedia PDF Downloads 166

Authors: Tahani Aljohani, Jialin Yu, Alexandra. I. Cristea

Abstract:

The more an educational system knows about a learner, the more personalised interaction it can provide, which leads to better learning. However, asking a learner directly is potentially disruptive, and often ignored by learners. Especially in the booming realm of MOOC Massive Online Learning platforms, only a very low percentage of users disclose demographic information about themselves. Thus, in this paper, we aim to predict learners’ demographic characteristics, by proposing an approach using linguistically motivated Deep Learning Architectures for Learner Profiling, particularly targeting gender prediction on a FutureLearn MOOC platform. Additionally, we tackle here the difficult problem of predicting the gender of learners based on their comments only – which are often available across MOOCs. The most common current approaches to text classification use the Long Short-Term Memory (LSTM) model, considering sentences as sequences. However, human language also has structures. In this research, rather than considering sentences as plain sequences, we hypothesise that higher semantic - and syntactic level sentence processing based on linguistics will render a richer representation. We thus evaluate, the traditional LSTM versus other bleeding edge models, which take into account syntactic structure, such as tree-structured LSTM, Stack-augmented Parser-Interpreter Neural Network (SPINN) and the Structure-Aware Tag Augmented model (SATA). Additionally, we explore using different word-level encoding functions. We have implemented these methods on Our MOOC dataset, which is the most performant one comparing with a public dataset on sentiment analysis that is further used as a cross-examining for the models' results.

Keywords: deep learning, data mining, gender predication, MOOCs

Procedia PDF Downloads 148

Authors: Gong Zhilin, Jing Yang, Jian Yin

Abstract:

The purpose of this paper is to develop a novel system for credit card fraud detection based on sequential modeling of data using hybrid deep learning models. The projected model encapsulates five major phases are pre-processing, imbalance-data handling, feature extraction, optimal feature selection, and fraud detection with an ensemble classifier. The collected raw data (input) is pre-processed to enhance the quality of the data through alleviation of the missing data, noisy data as well as null values. The pre-processed data are class imbalanced in nature, and therefore they are handled effectively with the K-means clustering-based SMOTE model. From the balanced class data, the most relevant features like improved Principal Component Analysis (PCA), statistical features (mean, median, standard deviation) and higher-order statistical features (skewness and kurtosis). Among the extracted features, the most optimal features are selected with the Self-improved Arithmetic Optimization Algorithm (SI-AOA). This SI-AOA model is the conceptual improvement of the standard Arithmetic Optimization Algorithm. The deep learning models like Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and optimized Quantum Deep Neural Network (QDNN). The LSTM and CNN are trained with the extracted optimal features. The outcomes from LSTM and CNN will enter as input to optimized QDNN that provides the final detection outcome. Since the QDNN is the ultimate detector, its weight function is fine-tuned with the Self-improved Arithmetic Optimization Algorithm (SI-AOA).

Keywords: credit card, data mining, fraud detection, money transactions

Procedia PDF Downloads 131

Authors: Xiuying Wang, Changliang Li, Bo Xu

Abstract:

A fixed-length feature vector is required for many machine learning algorithms in NLP field. Word embeddings have been very successful at learning lexical information. However, they cannot capture the compositional meaning of sentences, which prevents them from a deeper understanding of language. In this paper, we introduce a novel hierarchical tree long short-term memory (HTLSTM) model that learns vector representations for sentences of arbitrary syntactic type and length. We propose to split one sentence into three hierarchies: short phrase, long phrase and full sentence level. The HTLSTM model gives our algorithm the potential to fully consider the hierarchical information and long-term dependencies of language. We design the experiments on both English and Chinese corpus to evaluate our model on sentiment analysis task. And the results show that our model outperforms several existing state of the art approaches significantly.

Keywords: deep learning, hierarchical tree long short-term memory, sentence representation, sentiment analysis

Procedia PDF Downloads 349

Authors: Leah Lee

Abstract:

This paper examines the correlation between deep learning models and emotions with noises to see whether or not noises mask emotions. The deep learning models used are plain convolutional neural networks (CNN), auto-encoder, long short-term memory (LSTM), and Visual Geometry Group-16 (VGG-16). Emotion datasets used are Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), Crowd-sourced Emotional Multimodal Actors Dataset (CREMA-D), Toronto Emotional Speech Set (TESS), and Surrey Audio-Visual Expressed Emotion (SAVEE). To make it four times bigger, audio set files, stretch, and pitch augmentations are utilized. From the augmented datasets, five different features are extracted for inputs of the models. There are eight different emotions to be classified. Noise variations are white noise, dog barking, and cough sounds. The variation in the signal-to-noise ratio (SNR) is 0, 20, and 40. In summation, per a deep learning model, nine different sets with noise and SNR variations and just augmented audio files without any noises will be used in the experiment. To compare the results of the deep learning models, the accuracy and receiver operating characteristic (ROC) are checked.

Keywords: auto-encoder, convolutional neural networks, long short-term memory, speech emotion recognition, visual geometry group-16

Procedia PDF Downloads 75

Authors: Ali Kazemi

Abstract:

Those within the volatile and interconnected international economic markets, appropriately predicting market trends, hold substantial fees for traders and financial establishments. Traditional device mastering strategies have made full-size strides in forecasting marketplace movements; however, monetary data's complicated and networked nature calls for extra sophisticated processes. This observation offers a groundbreaking method for monetary marketplace prediction that leverages the synergistic capability of Graph Convolutional Networks (GCNs) and Long Short-Term Memory (LSTM) networks. Our suggested algorithm is meticulously designed to forecast the traits of inventory market indices and cryptocurrency costs, utilizing a comprehensive dataset spanning from January 1, 2015, to December 31, 2023. This era, marked by sizable volatility and transformation in financial markets, affords a solid basis for schooling and checking out our predictive version. Our algorithm integrates diverse facts to construct a dynamic economic graph that correctly reflects market intricacies. We meticulously collect opening, closing, and high and low costs daily for key inventory marketplace indices (e.g., S&P 500, NASDAQ) and widespread cryptocurrencies (e.g., Bitcoin, Ethereum), ensuring a holistic view of marketplace traits. Daily trading volumes are also incorporated to seize marketplace pastime and liquidity, providing critical insights into the market's shopping for and selling dynamics. Furthermore, recognizing the profound influence of the monetary surroundings on financial markets, we integrate critical macroeconomic signs with hobby fees, inflation rates, GDP increase, and unemployment costs into our model. Our GCN algorithm is adept at learning the relational patterns amongst specific financial devices represented as nodes in a comprehensive market graph. Edges in this graph encapsulate the relationships based totally on co-movement styles and sentiment correlations, enabling our version to grasp the complicated community of influences governing marketplace moves. Complementing this, our LSTM algorithm is trained on sequences of the spatial-temporal illustration discovered through the GCN, enriched with historic fee and extent records. This lets the LSTM seize and expect temporal marketplace developments accurately. Inside the complete assessment of our GCN-LSTM algorithm across the inventory marketplace and cryptocurrency datasets, the version confirmed advanced predictive accuracy and profitability compared to conventional and opportunity machine learning to know benchmarks. Specifically, the model performed a Mean Absolute Error (MAE) of 0.85%, indicating high precision in predicting day-by-day charge movements. The RMSE was recorded at 1.2%, underscoring the model's effectiveness in minimizing tremendous prediction mistakes, which is vital in volatile markets. Furthermore, when assessing the model's predictive performance on directional market movements, it achieved an accuracy rate of 78%, significantly outperforming the benchmark models, averaging an accuracy of 65%. This high degree of accuracy is instrumental for techniques that predict the course of price moves. This study showcases the efficacy of mixing graph-based totally and sequential deep learning knowledge in economic marketplace prediction and highlights the fee of a comprehensive, records-pushed evaluation framework. Our findings promise to revolutionize investment techniques and hazard management practices, offering investors and economic analysts a powerful device to navigate the complexities of cutting-edge economic markets.

Keywords: financial market prediction, graph convolutional networks (GCNs), long short-term memory (LSTM), cryptocurrency forecasting

Procedia PDF Downloads 66

Authors: Giovanni Cicceri, Roberta Maisano, Nathalie Morey, Salvatore Distefano

Abstract:

The main goal of this paper is to present a solution for a water purification system based on an Environmental Internet of Things (EIoT) platform to monitor and control water quality and machine learning (ML) models to support decision making and speed up the processes of purification of water. A real case study has been implemented by deploying an EIoT platform and a network of devices, called Gramb meters and belonging to the Gramb project, on wastewater purification systems located in Calabria, south of Italy. The data thus collected are used to control the wastewater quality, detect anomalies and predict the behaviour of the purification system. To this extent, three different statistical and machine learning models have been adopted and thus compared: Autoregressive Integrated Moving Average (ARIMA), Long Short Term Memory (LSTM) autoencoder, and Facebook Prophet (FP). The results demonstrated that the ML solution (LSTM) out-perform classical statistical approaches (ARIMA, FP), in terms of both accuracy, efficiency and effectiveness in monitoring and controlling the wastewater purification processes.

Keywords: environmental internet of things, EIoT, machine learning, anomaly detection, environment monitoring

Procedia PDF Downloads 151

Authors: Yin Yuanling

Abstract:

A Deep Feedback Recurrent Neural Network (DFRNN) and Bidirectional Long Short-Term Memory (BiLSTM) are designed to address the problem of low accuracy of traditional relationship extraction models. This method combines a deep feedback-based recurrent neural network (DFRNN) with a bi-directional long short-term memory (BiLSTM) approach. The method combines DFRNN, which extracts local features of text based on deep feedback recurrent mechanism, BiLSTM, which better extracts global features of text, and Self-Attention, which extracts semantic information. Experiments show that the method achieves an F1 value of 76.69% on the CEC dataset, which is 0.0652 better than the BiLSTM+Self-ATT model, thus optimizing the performance of the deep learning method in the event relationship extraction task.

Keywords: event relations, deep learning, DFRNN models, bi-directional long and short-term memory networks

Procedia PDF Downloads 144