Neural Network Learning Algorithm Research Articles

Comparison of Conventional Machine Learning and Deep Neural Network Algorithms in the Prediction of Monkey-Pox

Smallpox syndrome, or monkeypox, is an uncommon zoonotic viral infection brought on by the monkeypox virus, which belongs to the genus Orthopoxvirus and family Poxviridae. Injury-related mortality in primates ranges from 1 to 10%. Data mining is a method for analyzing data. Deep neural networks and traditional machine learning methods are used in data analysis. The Python programming language is used during the comparison procedure of this research algorithm to generate values for accuracy, f1 score, precision, recall, ROC, and AUC. The test results demonstrate that using sigmoid activation function parameters, the deep neural network algorithm's accuracy is 70.08%, F1 score is 79.18%, precision is 68.59%, recall is 62.65%, and AUC is 62.65%. Compared to conventional machine learning algorithms, the adagrad optimizer has a higher value with a learning rate of 0.01 and 0.2 dropouts. The conventional machine learning model algorithm has the best xgboost, F1 score, precision, recall, and AUC scores compared to other approaches: 64.40%, 64.45%, and 78.14%. According to these numbers, the average fairness disparity between deep neural network algorithms and traditional machine learning is 5.68%, F1 score is 13.79%, precision is 4.14%, recall is 1.75%, and AUC is 1.75%.

Machine learning algorithm for an artificial neural network for building a model of managerial decision-making when developing a marketing strategy

The article describes the application of the machine learning method, namely the backpropagation algorithm, in order to optimize managerial decision making when developing a marketing strategy. A qualitative analysis of data processing has been carried out, which proves the relevance of using the backpropagation method in marketing interpretation. Using the example of a task with the choice of hashtags for social media, a five-step model has been built step by step, which, after passing through many iterations of machine learning algorithms, could automate the solution of problems similar to the original one. The step-by-step process of machine learning has been described in terms of logic, mathematical functioning and programming. KPIs to assess the accuracy of the task by the model have been highlighted. An example of comparing the base KPIs of two models to select a more accurate one has been given

Comparison of Conventional Machine Learning and Deep Neural Network Algorithms in the Prediction of Monkey-Pox

Smallpox syndrome, also known as monkeypox, is an uncommon zoonotic viral infection brought on by the monkeypox virus, which belongs to the genus orthopoxvirus and family Poxviridae. Injury-related mortality in primates ranges from 1 to 10%. Data mining is a method for analyzing data. Deep neural networks and traditional machine learning methods are both used in the data analysis process. The Python programming language is used during the comparison procedure of this research algorithm to generate values for accuracy, f1 score, precision, recall, ROC, and AUC. The test results demonstrate that using sigmoid activation function parameters, the deep neural network algorithm's accuracy is 70.08%, F1 score is 79.18%, precision is 68.59%, recall is 62.65%, and AUC is 62.65%. In comparison to using conventional machine learning algorithms, the adagrad optimizer with learning rate 0.01 and 0.2 dropout has a higher value. The conventional machine learning model algorithm has the best xgboost, F1 score, precision, recall, and AUC scores when compared to other approaches: 64.40%, 64.45%, and 78.14%. According to these numbers, the average fairness disparity between deep neural network algorithms and traditional machine learning is 5.68%, F1 score is 13.79%, precision is 4.14%, recall is 1.75%, and AUC is 1.75%.

Combining Reinforcement Learning Algorithms with Graph Neural Networks to Solve Dynamic Job Shop Scheduling Problems

Smart factories have attracted a lot of attention from scholars for intelligent scheduling problems due to the complexity and dynamics of their production processes. The dynamic job shop scheduling problem (DJSP), as one of the intelligent scheduling problems, aims to make an optimized scheduling decision sequence based on the real-time dynamic job shop environment. The traditional reinforcement learning (RL) method converts the scheduling problem with a Markov process and combines its own reward method to obtain scheduling sequences in different real-time shop states. However, the definition of shop states often relies on the scheduling experience of the model constructor, which undoubtedly affects the optimization capability of the reinforcement learning model. In this paper, we combine graph neural network (GNN) and deep reinforcement learning (DRL) algorithm to solve DJSP. An agent model from job shop state analysis graph to scheduling rules is constructed, thus avoiding the problem that traditional reinforcement learning methods rely on scheduling experience to artificially set the state feature vectors. In addition, a new reward function is defined, and the experimental results prove that our proposed reward method is more effective. The effectiveness and feasibility of our model is demonstrated by comparing with general deep reinforcement learning algorithms on minimizing the earlier and later completion time, which also lays the foundation for solving the DJSP later.

Comparing Mechanical Neural-Network Learning Algorithms

Abstract The purpose of this work is to compare learning algorithms to identify which is the fastest and most accurate for training mechanical neural networks (MNNs). MNNs are a unique class of lattice-based artificial intelligence (AI) architected materials that learn their mechanical behaviors with repeated exposure to external loads. They can learn multiple behaviors simultaneously in situ and re-learn desired behaviors after being damaged or cut into new shapes. MNNs learn by tuning the stiffnesses of their constituent beams similar to how artificial neural networks (ANNs) learn by tuning their weights. In this work, we compare the performance of six algorithms (i.e., genetic algorithm, full pattern search, partial pattern search, interior point, sequential quadratic progression, and Nelder–Mead) applied to MNN leaning. A computational model was created to simulate MNN learning using these algorithms with experimentally measured noise included. A total of 3900 runs were simulated. The results were validated using experimentally collected data from a physical MNN. We identify algorithms like Nelder–Mead that are both fast and able to reject noise. Additionally, we provide insights into selecting learning algorithms based on the desired balance between accuracy and speed, as well as the general characteristics that are favorable for training MNNs. These insights will promote more efficient MNN learning and will provide a foundation for future algorithm development.

Seismic Sedimentologic Study on Sequence-Sediment Evolution in Fault Basin Regions

With potential hydrocarbon source rocks, the Mesozoic Lower Cretaceous interval in the Aitegle Sag of the Yingen-Ejinaqi Basin in China has a bright petroleum exploration prospect. Considering the scarcity of wells in Aitegle Sag, we have systematically carried out the research on the sequence structure and sedimentary system evolution of the Lower Cretaceous based on seismic, core, and logging data, guided by the Vail classic sequence stratigraphy, and innovatively introduced the self-organizing mapping neural network (SOM) learning algorithm to assist in seismic geological analysis. The results are as follows: (1) after RMSAmp, LyapIndex, FrctDim, and InfoEntr clustering, five zones are divided, which are corresponding to five seismic facies (showing medium-strong amplitude and parallel-subparallel reflection; medium-strong amplitude and wedge-shaped divergent reflection; medium-strong amplitude and continuous-relatively continuous, parallel-subparallel reflection; strong amplitude and continuous sheet-like or blank reflection; and weak amplitude and mound-shaped chaotic reflection, respectively), and interpreted as five subfacies including delta plain front, prodelta-shallow lake, semideep lake, deep lake, and sublacustrine fan. (2) The Lower Cretaceous formation includes one second-order sequence, three third-order sequences (SQ2–SQ4), and six system tracts. In the SQ2–SQ4 circle, the delta system formed by sediments from southeast varies remarkably in the progradational zones in different system tracts, and tectonic movements and sediment supply control the sequence stratigraphy in the study area as major factors. (3) It is feasible to apply SOM learning algorithm to seismic attribute clustering analysis. This method effectively realizes the combination of geophysical methods based on big data analysis and geological problem analysis, which not only strengthens the use of seismic data but also ensures the effectiveness and reliability of clustering results. These findings offer basic data and scientific supports for the exploration and development of unconventional reservoirs (e.g., shale and tight oil/gas reservoirs) in the Aitegle Sag and also provide new ideas for studying the sequence-sediment in the target zones in the fault basin regions with scarce wells.

BP-based supervised learning algorithm for multilayer photonic spiking neural network and hardware implementation.

We introduce a supervised learning algorithm for photonic spiking neural network (SNN) based on back propagation. For the supervised learning algorithm, the information is encoded into spike trains with different strength, and the SNN is trained according to different patterns composed of different spike numbers of the output neurons. Furthermore, the classification task is performed numerically and experimentally based on the supervised learning algorithm in the SNN. The SNN is composed of photonic spiking neuron based on vertical-cavity surface-emitting laser which is functionally similar to leaky-integrate and fire neuron. The results prove the demonstration of the algorithm implementation on hardware. To seek ultra-low power consumption and ultra-low delay, it is great significance to design and implement a hardware-friendly learning algorithm of photonic neural networks and realize hardware-algorithm collaborative computing.

Deep Learning–Based Estimation of Implantable Collamer Lens Vault Using Optical Coherence Tomography

To develop and validate a deep learning neural network for automated measurement of implantable collamer lens (ICL) vault using anterior segment optical coherence tomography (AS-OCT). Cross-sectional retrospective study. Multicentric. A total of 2,647 AS-OCT anonymized eye scans were used from 139 eyes of 82 subjects who underwent ICL surgery in 3 different centers. Using transfer learning, a deep learning neural network algorithm was trained and validated for the regression task of estimating the ICL vault on the MS-39 OCT (CSO, Florence, Italy). A trained operator separately reviewed all OCT scans and measured the central vault using a built-in OCT caliper tool. The model was then separately tested on 191 AS-OCT scans. A Bland-Altman plot was constructed and the mean absolute percentage error (MAPE), the mean absolute error (MAE), the root mean squared error (RMSE), the Pearson correlation coefficient (r), and the determination coefficient (R2) were calculated to evaluate the strength and validity of the model. On the test set, the model achieved a MAPE of 3.42%, an MAE of 15.82 µm, an RMSE of 18.85 µm, a Pearson correlation coefficient r of +0.98 (P-Value < .00001), and a coefficient of determination R2 of +0.96. There was no significant difference between the vaults of the test set labeled by the technician versus those estimated by the model: 478 ± 95 µm versus 475 ± 97 µm, respectively, P-Value = 0.064). Using transfer learning, our deep learning neural network was able to accurately compute the ICL vault from AS-OCT scans, overcoming the limitations of an imbalanced dataset and limited training data. Such an algorithm can assist the postoperative assessment in ICL surgery. PRéCIS: This study develops and validates a deep learning model to accurately estimate implantable collamer lens vault on optical coherence tomography using transfer learning, convoluted neural networks, and regression.

The Comparison of Long Short-Term Memory Neural Network and Deep Forest for the Evaporation Duct Height Prediction

An evaporation duct is a type of atmospheric stratification that affects radio systems. Atmospheric duct prediction is helpful for radar detection. In this paper, we used the deep forest, which is different from a deep learning framework, to predict the atmospheric duct height. At the same time, the long short-term memory neural network and other machine learning algorithms, such as the logistic regression, random forest, Bayes, and support vector regression algorithms, were adopted to predict the evaporation duct height. The predicted results with filled and unfilled missing data show that an accurate prediction of the evaporation duct height can be achieved using the deep forest.

Pengembangan Sistem Prediksi Bantuan Program Keluarga Harapan (PKH) Berbasis Machine Learning

The Family Hope Program (PKH) is a poverty alleviation program which is one of the government's strategies in reducing the poverty line. This program provides cash social assistance to poor families who are included in the list of beneficiary families with a focus on education and health. The purpose of implementing the PKH program is not only to reduce poverty and increase human resources but to break the poverty chain. The implementation of PKH in its realization experienced many obstacles that caused the program not to be on target, this was because the data verification process was not yet effective and was still carried out manually. A process is needed to digitize the distribution and realization of the family of hope program. Through this research, a system was developed that can predict the value of PKH beneficiary assistance. The system developed is based on machine learning with a prediction model using Artificial Neural Network (ANN) and Backpropagation learning algorithm. Parameters in the learning system using PKH assessment as many as 8 indicators from the data of PKH beneficiaries in Tabanan Regency. Based on the prediction model testing using two data treatments, namely with and without preprocessing data. Parameters treated with data on numeric attributes and categories provide optimal values with an R2 Score of 0.695824 with a number of hidden layers of 500 and a max epoch of 375

Machine learning algorithms in wood ash-cement-Nano TiO2-based mortar subjected to elevated temperatures

Mortar is subjected to high temperatures during fire attacks or when it is near heat-radiating equipment like furnaces and reactors. The physical and microstructure of mortar were considerably altered by high temperatures. In this investigation, the effects of elevated temperatures on the flexural and compressive strengths of wood ash (WA) cement mortar modified with green-synthesised Nano titanium oxide (NT) were examined. In order to produce mortar samples, the cement was replaced with 10% WA, and 1–3% NT by weight of binder were added at constant water-binder ratio. The specimens were heated to 105, 200, 400, 600, and 800 °C with an incremental rate of 10 °C per min in the electric furnace for a sustained period of 2 h to measure their strengths. The machine learning algorithm of artificial neural networks with Levenberg-Marquardt backpropagation training techniques of different network architectures was engaged to predict the compressive strength of WA-cement-NT-based mortar produced. The findings showed that higher temperatures reduced compressive strength after 400 °C and flexural strength after 200 °C. The mortar specimen with a 3% NT addition showed the highest residual compressive strength increase, ranging from 18.75 to 27.38%. Compared to compressive strength, flexural strength is more severely affected by high temperatures. The backpropagation training algorithm revealed that each hidden layer displayed its unique strong prediction. However, Levenberg-Marquardt backpropagation training technique of 7–10-10-1 network structures yielded the best performance metrics for training, validation, and testing compared to 7-10-10-10 and 7-10-1 network architectures.

Source tracing of tree smoke in forest fires based on laser-induced breakdown spectroscopy

The damage caused by forest fires to humans and the environment cannot be ignored. However, there are few works about the traceability of tree smoke in current time. In this paper, a new system of laser-induced breakdown spectroscopy (LIBS) combined with machine learning was developed to detect and identify the smoke from different tree species. In this new system, spectral lines of heavy elements including Fe, Sr and Ba can be observed, which, together with spectral lines of some other common elements, a total of 57 lines, are regarded as characteristic fingerprints of smoke spectra. These spectral lines are used as variables to reduce the dimensionality of the data using principal component analysis (PCA). Then, a supervised machine learning algorithm of back propagation artificial neural networks (BP-ANN) are applied to identify the source of tree smoke. The average precision rate for smoke achieved over 85%, indicating that LIBS combined with machine learning has a wide application prospect in situ online detection and identification of local forest fires.

GDP responses to supply chain disruptions in a post-pandemic era: Combination of DL and ANN outputs based on Google Trends

With the recent Russian-Ukraine conflict, the frequency and intensity of disruptive shocks on major supply chains have risen, causing increasing food and energy security concerns for regulators. That is, the combination of newly available sophisticated deep learning tools with real-time series data may represent a fruitful policy direction because machines can identify patterns without being pre-conditioned calibration thanks to experimental data training. This paper employs Deep Learning (DL) and Artificial Neural Network (ANN) algorithms and aimed predicts GDP responses to supply chain disruptions, energy prices, economic policy uncertainty, and google trend in the US. Sampled data from 2008 to 2022 are monthly wrangled and embed different recession episodes connected to the subprime crisis of 2008, the COVID-19 pandemic, the recent invasion of Ukraine by Russia, and the current economic recession in the US. Both DL and ANN outputs empirically (and unanimously) demonstrated how sensitive monthly GDP variations are to dynamic changes in supply chain performances. Findings identify the substantial role of google trends in delivering a consistent fit to predicted GDP values, which has implications While a comparative discussion over the larger forecasting performance of DL compared to ANN experiments is offered, implications for global policy, decision-makers and firm managers are finally provided.

Identification of Insulator Contamination Status Based on Multi-spectral Image Fusion and Radial Basis Function Neural Network

Insulator contamination monitoring is an important way to avoid insulator contamination and maintain insulation performance. In order to ensure power supply and achieve contactless detection of insulator contamination status, a method is proposed in this paper to identify insulator contamination status by adopting infrared, ultraviolet, and visible multi-spectral image information fusion. An insulator with different contamination states in a number of substations is taken as the research object in this paper. The image segmentation is performed by using the seed region growth method to extract the infrared, ultraviolet and visible features of the insulator surface, and the radial basis function neural network learning algorithm is used to classify and decompose and fuse the images according to their different local area energies. The comparison of the recognition rates using infrared and ultraviolet features with those fused shows that the method has significant advantages and provides a new method for the detection of insulator contamination status.

A hybrid machine learning and embedded IoT-based water quality monitoring system

Since access to clean and safe water is one of life's most basic needs, the exponential growth of the world's population makes it vital to guarantee a workable framework. Manually collecting samples and sending them to a research center for discovery and analysis is the typical approach for gathering information on water qualities. However, this approach is illogical in the long run because it requires a lot of time and labor. This paper aimed to construct, test, and evaluate the usefulness of machine learning and the Internet of Things (IoT) at water storage stations. First, we developed a system prototype and assessed its performance using classifier and reliability matrices. This study considers water's physical and chemical parameters to evaluate the level of water pollutants present in drinking water. The parameters measured include temperature, pH, turbidity, Dissolved Oxygen (DO), Total Dissolved Solids (TDS), Oxidation Reduction Potential (ORP), and electrical conductivity. After analyzing the sensor data, we used Artificial Neural Network (ANN) and Support Vector Machine (SVM) machine learning algorithms to forecast the impurity level of the water measured. The performance showed that the ANN models used have the highest accuracy and are the most suitable to predict water source and status. We also introduced a water treatment method to provide an automated corrective measure based on a specific amount of water contamination. Based on the system's results, we concluded that AI and IoT are more efficient in remotely monitoring safe and harmful water conditions.

Territorial spatial vulnerability assessment based on PSO-BP neural network: A case study in Shenzhen, China

Studies on vulnerability, a fairly new research paradigm inspired by global environmental change, are expanding to support scientific decision-making for developing sustainable, resilient territorial space at the regional level. The goal of this research is to reveal the vulnerability of a human-environment coupled system and its interaction mechanisms amidst a changing environment. This study started with the vulnerability scoping diagram (VSD) conceptual model and built up a research framework of territorial spatial vulnerability according to the three structural elements of ecological protection, agricultural production and urban construction. Following the requirements for intelligent computing, we introduced particle swarm optimization and BP neural network learning algorithms to evaluate the vulnerability of Shenzhen's territory. We also introduced flow cytometry to analyze the mechanism of Shenzhen's territorial spatial vulnerability driven by human and natural forces. The results showed that the territorial spatial vulnerability of Shenzhen was low under various functional orientations. Among them, ecological protection-oriented territorial spatial vulnerability was high in the west and low in the east and followed an inverted U-shaped trend in the north-south direction. Agricultural production-oriented territorial spatial vulnerability was high in the east and low in the west, high in the north and low in the south. Urban construction-oriented territorial spatial vulnerability was high in the west and low in the east, high in the north and low in the south. Under the complex orientation of the gigantic human-environment coupled system, the vulnerability of territory was symmetrical and balanced in the north-south direction and fluctuated with an overall declining trend in the east-west direction. In vulnerability, ecological protection, agricultural production and urban construction systems independently accounted for 35.642%, 38.209%, and 26.149% of the territorial spatial vulnerability, respectively.

SSI-LSTM network for adaptive operational modal analysis of building structures

Various operational modal analysis (OMA) methods have been developed to identify the modal parameters of buildings in use. Recently, efforts have been expended to solve the existing problems associated with accuracy and computational time of OMA methods using deep learning algorithms of convolution neural network (CNN) and deep neural network (DNN). Considering that the result of modal parameters involves a process associated with the extraction of the dynamic characteristics by separating the periodic modes based on a complex ambient vibration response, the neural network technique, whereby the outputs of the past and present data are independent and static, may not be suitable for application in OMA. In this study, an adaptive, stochastic subspace identification long-short term memory (SSI-LSTM) method is proposed to evaluate the variations of the modal parameters as an indicator of the dynamic characteristics of a structure in the time history data of the structural response measured in real-time. Both past and present responses can be considered in the adaptive SSI-LSTM method and can estimate the natural frequency that changes with aging and damage to buildings. Simultaneously, as the deep learning method is basically used, the problems of accuracy and computational time continuously raised in the OMA method can be solved. The proposed method can configure the OMA method using the modal parameter results from the ambient vibration response and SSI for direct network learning. The covariance-driven SSI (SSI-COV) method is used to construct the training data for the proposed adaptive SSI-LSTM method. The proposed SSI-LSTM method is verified by applying it to a simulation model with four degrees-of-freedom and a three-story experimental model. Lastly, the practicality of the proposed method is validated using the response obtained from a 55-story high-rise building.

Machine learning based ground motion site amplification prediction

Site condition impact on seismic ground motion has been a complex but important subject in earthquake hazard analysis. Traditional studies on site amplification effect are either based on site response via wave propagation simulation or regression analysis using parameters such as Vs30, bedrock ground motion and site response period. Ground Motion Prediction Equations (GMPEs) are used for regions where there is limited data of seismic records. The main issues with these approaches are that they cannot demonstrate the complex relationship between site amplification and its various affecting parameters, thus there exists large uncertainty in the results. Recent studies based on machine learning have shown significant improvement in predicting the site amplification, but the result is not well explained. This study assembled the information on 6 parameters including Vs30, magnitude, epicentral distance, earthquake source depth, bedrock ground motion, and altitude of 353,327 records observed during 1997 and 2019 from 698 KiK-net stations. Three machine learning algorithms of Random Forest (RF), XGBoost, and Deep Neural Networks (DNN) were implemented to predict the site amplification factor using these 6 selected parameters. Shapley Additive explanation (SHAP) was used to explain the importance of the 6 parameters. The results show that all three machine learning algorithms performed much better than the traditional GMPE approach with XGBoost’s performance the best. The explanation provided by the SHAP analysis further enhanced the reasonability of this study. It is anticipated that the combination of machine learning and SHAP analysis can provide better assessment for site amplification of ground motion with better potential of future application in seismic hazard analysis.

Hybrid spam message detection using convolutional neural network and long short-term memory techniques

Short Message Service (SMS) is a feature of a mobile phone that enable convenient and instant way of sending electronic messages between users. As SMS usage increases fraudulent text messages, known as spam, are becoming more common. Spam SMS may result in leaking personal information, invasion of privacy or accessing unauthorized data from mobile devices. Users of mobile phones can mistakingly give away personal information with the assumption that they are sharing it with the right recipients. This work propose a SMS spam detection method that combines convolutional neural network (CNN) and long short term memory (LSTM) deep learning algorithms. The CNN is used for feature extraction while the LSTM classifies the message. The SMS spam dataset, collected from online repository, is used to train the model. Word embeddings is used to vectorize the words in the message to make it suitable for the model. The result obtained from the implementation outperforms other machine learning algorithms with an accuracy of 99.77%.

The Research of Short-term Electric Load Forecasting based on Machine Learning Algorithm

The ability of electric load forecasting become the key to measuring electric planning and dispatching, and improving the accuracy of electric load forecasting has become one hot spot topic of scholars in recent years. The traditional electric load forecasting algorithm mainly include statistical learning algorithm. The electric load forecasting algorithm based on traditional forecasting algorithm, machine learning algorithm and neural network algorithm greatly improve the convergence approximation effect and accuracy. In recent years, the electric load forecasting algorithms based on the deep learning algorithm has greatly reduced the error and improved the measurement accuracy and robustness, such as RNN, GRU, LSTM, DBN, TCN, and so on, as well as the combination algorithm based on deep learning. This paper introduces the classical algorithm of deep learning in electric load forecasting. The main purpose is to explore the error and fitting degree of algorithms established by different algorithms, and provide reference for the selection of forecasting algorithms for various types of electric load forecasting.