LSTM Methods Research Articles

Performance Comparison of ARIMA, LSTM, and Prophet Methods in Sales Forecasting

The development of the business world that is growing rapidly today resulted in tighter competitiveness between fellow business actors. One of the businesses that has sprung up in the market today is the bakery business. Currently, bread is one of the food needs in Indonesia that is great demand by children to the elderly, which is often used as breakfast or snack. One of the companies that produces white bread is the Bandung White Bread Factory. The number of sales at this factory continues to increase every month based on total sales data recorded since 2021. With the increasing number of sales at this factory, the factory often experiences stock shortages and cannot meet customer demand. Therefore, in this study, a model has been developed to forecast the sales of white bread using the ARIMA, LSTM, and Prophet methods. The results of the study showed that the ARIMA method (1,0,2) had the best performance compared to the LSTM and Prophet methods, because the ARIMA method (1,0,2) produced the smallest error accuracy value, namely with a MAPE value of 4.548%, an MSE value of 2248.0822, and an RMSE value of 47.4139.

Performance Analysis and Prediction of 5G Round-Trip Time Based on the VMD-LSTM Method

With the increasing level of industrial informatization, massive industrial data require real-time and high-fidelity wireless transmission. Although some industrial wireless network protocols have been designed over the last few decades, most of them have limited coverage and narrow bandwidth. They cannot always ensure the certainty of information transmission, making it especially difficult to meet the requirements of low latency in industrial manufacturing fields. The 5G technology is characterized by a high transmission rate and low latency; therefore, it has good prospects in industrial applications. To apply 5G technology to factory environments with low latency requirements for data transmission, in this study, we analyze the statistical performance of the round-trip time (RTT) in a 5G-R15 communication system. The results indicate that the average value of 5G RTT is about 11 ms, which is less than the 25 ms of WIA-FA. We then consider 5G RTT data as a group of time series, utilizing the augmented Dickey–Fuller (ADF) test method to analyze the stability of the RTT data. We conclude that the RTT data are non-stationary. Therefore, firstly, the original 5G RTT series are subjected to first-order differencing to obtain differential sequences with stronger stationarity. Then, a time series analysis-based variational mode decomposition–long short-term memory (VMD-LSTM) method is proposed to separately predict each differential sequence. Finally, the predicted results are subjected to inverse difference to obtain the predicted value of 5G RTT, and a predictive error of 4.481% indicates that the method performs better than LSTM and other methods. The prediction results could be used to evaluate network performance based on business requirements, reduce the impact of instruction packet loss, and improve the robustness of control algorithms. The proposed early warning accuracy metrics for control issues can also be used to indicate when to retrain the model and to indicate the setting of the control cycle. The field of industrial control, especially in the manufacturing industry, which requires low latency, will benefit from this analysis. It should be noted that the above analysis and prediction methods are also applicable to the R16 and R17 versions.

Long Short-Term Memory and Gated Recurrent Unit Modeling for Stock Price Forecasting

The rapid advancement of deep learning technology offers significant benefits, particularly for time series data forecasting. LSTM and GRU are two deep learning methods used for forecasting. This study aims to compare the LSTM and GRU methods in predicting the stock prices of PT Mayora Indah Tbk, listed on the Indonesia Stock Exchange (IDX), accessed through Yahoo Finance. The model architecture was developed using combinations of data splitting, learning rate, epoch count, and neuron count. The combinations used in this study include data splits of 70:30 and 80:20, with varying parameter combinations of learning rates at 0.01, 0.001, and 0.0001, epoch counts of 50, 100, and 150, and neuron counts of 50 and 100. The results indicate that the GRU method outperforms the LSTM method in prediction accuracy. The RMSE and MAPE tests show that the GRU method yields RMSE and MAPE of 34.4233 and 1.27%, respectively, compared to the LSTM method with RMSE and MAPE of 35.3775 and 1.28%. The best GRU architecture was achieved with a learning rate of 0.001, 50 neurons, and 150 epochs, while the best LSTM architecture was achieved with a learning rate of 0.001, 100 neurons, and 150 epochs. The optimal architecture for both methods was found with a data split of 70:30

Optimization of Marketing Strategy for State-Owned Energy Products through Sentiment Analysis with VADER and LSTM on Social Media

Sentiment analysis is also known as opinion mining. It has an important role in natural language processing and data mining. It involves extracting and analyzing subjective information from textual data to determine the sentiment. With the advancement of technology, it is increasingly important to understand users' opinions and sentiments regarding a particular product, service or issue. This research aims to optimize the marketing strategy of energy products in SOE subsidiaries through sentiment analysis using the VADER and LSTM methods on social media.The researcher analyzed data from twitter to identify consumer sentiment towards energy products. The results of the analysis show that the combination of VADER and LSTM is effective in identifying the nuances of sentiment, with the accuracy of sentiment classification reaching 85%. Based on the analysis, the researchers developed recommendations for marketing strategies that are more focused and responsive to public perception. This study also highlights the importance of two-way interaction between companies and consumers as part of an effective marketing strategy. The findings of this study are expected to be used as a reference by state-owned companies in the energy sector to improve the effectiveness of their marketing strategies through a sentiment analysis approach on social media.

An improved convolutional recurrent neural network for stock price forecasting

<p>Stock price forecasting is a challenging area of research, particularly due to the complexity and unpredictability of financial markets. The accuracy of prediction models is influenced by various factors, including nonlinearity, seasonality, and economic shocks. Deep learning has demonstrated better forecasts of stock prices than traditional approaches. This study, therefore, proposed a new approach to improve forecasting system based on an end-to-end convolutional recurrent neural network (CRNN) with attention mechanism. Our approach first investigates local stock price features using 1D convolutional neural network, and then employs a bidirectional long short-term memory (Bi-LSTM) network for forecasting. This model stands out by effectively utilizing contextual data and representing the temporal character of data. The Bi-LSTM is helpful for understanding the history and future contextual information since it uncovers both past and future contexts of stock data. Furthermore, integrating attention mechanism within the CRNN represents a significant improvement. This allows our model to handle long input sequences more effectively and capture the inherent stochasticity in stock prices, which is often missed by traditional models. The effectiveness of our approach is investigated using data on 10 stock indexes from Yahoo Finance. The results show that our method outperforms ARIMA, LSTM, and conventional methods.</p><p> </p>

Sentiment Analysis of Alfagift Application User Reviews Using Long Short-Term Memory (LSTM) and Support Vector Machine (SVM) Methods

The rapid advancement of mobile apps has emerged as an important aspect of the routine of internet-connected users. In Indonesia, many companies are introducing their apps to improve the quality of service for users, and Alfamart is one of them. However, users have identified many shortcomings in these apps. This feedback is provided by users on the review feature of the Alfagift app on the Google Play Store. This research aims to apply a sentiment analysis approach to identify the application's shortcomings so that developers can understand the aspects that need to be improved to improve the quality of application services. The research stages include data collection, preprocessing, labeling, weighting, classification of LSTM and SVM methods, and performance evaluation using a confusion matrix. The dataset consists of 1000 reviews obtained through web scraping techniques. This research uses the Lexicon-based method to classify the dataset into positive, negative, and neutral categories. The analysis results show that 801 data are classified as positive sentiment, 77 as negative sentiment, and 122 as neutral sentiment. Based on testing, both SVM and LSTM methods show good performance. The best accuracy results were obtained using the SVM method, which amounted to 83.5%. Meanwhile, the LSTM method achieved an accuracy of 82%.

Implementation of Text Mining to Detect Emotions of Fuel Price Increase Using BERT-LSTM Method

Fuel is essential in everyday life to be used as a means of transportation, which generally uses fuel oil. In early April 2022, there was an issue that caused an uproar in Indonesian society: rising fuel prices. Based on the problems described, this research uses emotion classification with the BERT and LSTM methods using social media data, such as YouTube, to classify types of emotions based on community responses to government decisions. The highest results were obtained when oversampling was used, with an accuracy of 95%, while without oversampling, the accuracy was only 47%. In the experiments that have been carried out, it turns out that using the oversampling method and the BERT method for emotion classification will cause a reduction in accuracy for the test section.

Selection of Modelling for Forecasting Crude Palm Oil Prices Using Deep Learning (GRU & LSTM)

The unstable crude palm oil (CPO) prices have an impact on assessments of economic growth and environmental sustainability, as well as market strategies, international trade discussions, and consumer pricing expectations for products made from CPO. Therefore, it is crucial to identify the best prediction method to accurately forecast this cost. This research aims to develop an accurate time series data prediction model for crude palm oil prices using GRU and LSTM methods. The study also aims to identify the best-performing model by comparing their performance. This study uses LSTM and GRU methods as well as Bi-LSTM as a comparison, using crude palm oil price data from the Indonesian Ministry of Trade (August 1, 2018–August 31, 2023) from Medan (1064 data points) and Rotterdam SPOT (617 data), Rotterdam Forward 1 Month (1,022 data), and Rotterdam Forward 2 Month (950 data). Each dataset is then split into training and testing data with a 70:30 ratio. The hyperparameters used are (a) learning rate: (0.001); (b) batch size: 100; (c) node: (512); (d) optimizer: Adam; (e) and epoch: 50. The results of the forecast are highly accurate with a MAPE of less than 10%. Overall, both LSTM and GRU techniques demonstrate excellent performance in forecasting crude palm oil prices, but they may need to be modified based on data features. The ARIMA method can also be considered for forecasting. Future studies may consider optimizing parameters and model structures. Doi: 10.28991/ESJ-2024-08-03-05 Full Text: PDF

Analysis and forecasting of syphilis trends in mainland China based on hybrid time series models.

Syphilis remains a serious public health problem in mainland China that requires attention, modelling to describe and predict its prevalence patterns can help the government to develop more scientific interventions. The seasonal autoregressive integrated moving average (SARIMA) model, long short-term memory network (LSTM) model, hybrid SARIMA-LSTM model, and hybrid SARIMA-nonlinear auto-regressive models with exogenous inputs (SARIMA-NARX) model were used to simulate the time series data of the syphilis incidence from January 2004 to November 2023 respectively. Compared to the SARIMA, LSTM, and SARIMA-LSTM models, the median absolute deviation (MAD) value of the SARIMA-NARX model decreases by 352.69%, 4.98%, and 3.73%, respectively. The mean absolute percentage error (MAPE) value decreases by 73.7%, 23.46%, and 13.06%, respectively. The root mean square error (RMSE) value decreases by 68.02%, 26.68%, and 23.78%, respectively. The mean absolute error (MAE) value decreases by 70.90%, 23.00%, and 21.80%, respectively. The hybrid SARIMA-NARX and SARIMA-LSTM methods predict syphilis cases more accurately than the basic SARIMA and LSTM methods, so that can be used for governments to develop long-term syphilis prevention and control programs. In addition, the predicted cases still maintain a fairly high level of incidence, so there is an urgent need to develop more comprehensive prevention strategies.

Leading Point Multi-Regression Model for Detection of Anomalous Days in German Energy System

In this article, the Leading Point Multi-Regression model was applied to identify days with anomalous energy consumption profiles. The data for the analysis come from the German energy system and they represent the hourly energy demand observed between 2006 and 2015. Days with abnormal daily profiles were identified based on the statistical analysis of the errors observed for the model. The accuracy of the model is very high and comparable with other models, e.g., LSTM, K-means, Recurrent NN, and tree-based ML methods. However, these methods rely on external factors (e.g., humidity, temperature, and sunshine) impacting energy consumption while our model uses only the energy consumption at specific fixed hours, regardless of external factors, thus being universal. Days with anomalous energy consumption profiles were identified as days related to celebration of New Year’s Eve and the New Year. Also, anomalies were identified for some other days, which were not that obvious, including Good Friday, National Day of Mourning, and, interestingly, the day of the Germany–Turkey match during the European Championship in 2008.

From simple to complex: a sequential method for enhancing time series forecasting with deep learning

Abstract Time series forecasting is a well-known deep learning application field in which previous data are used to predict the future behavior of the series. Recently, several deep learning approaches have been proposed in which several nonlinear functions are applied to the input to obtain the output. In this paper, we introduce a novel method to improve the performance of deep learning models in time series forecasting. This method divides the model into hierarchies or levels from simpler to more complex ones. Simpler levels handle smoothed versions of the input, whereas the most complex level processes the original time series. This method follows the human learning process where general/simpler tasks are performed first, and afterward, more precise/harder ones are accomplished. Our proposed methodology has been applied to the LSTM architecture, showing remarkable performance in various time series. In addition, a comparison is reported including a standard LSTM and novel methods such as DeepAR, Temporal Fusion Transformer, NBEATS and Echo State Network.

Diagnostic biomarker discovery from brain EEG data using LSTM, reservoir-SNN, and NeuCube methods in a pilot study comparing epilepsy and migraine

The study introduces a new online spike encoding algorithm for spiking neural networks (SNN) and suggests new methods for learning and identifying diagnostic biomarkers using three prominent deep learning neural network models: deep BiLSTM, reservoir SNN, and NeuCube. EEG data from datasets related to epilepsy, migraine, and healthy subjects are employed. Results reveal that BiLSTM hidden neurons capture biological significance, while reservoir SNN activities and NeuCube spiking dynamics identify EEG channels as diagnostic biomarkers. BiLSTM and reservoir SNN achieve 90 and 85% classification accuracy, while NeuCube achieves 97%, all methods pinpointing potential biomarkers like T6, F7, C4, and F8. The research bears implications for refining online EEG classification, analysis, and early brain state diagnosis, enhancing AI models with interpretability and discovery. The proposed techniques hold promise for streamlined brain-computer interfaces and clinical applications, representing a significant advancement in pattern discovery across the three most popular neural network methods for addressing a crucial problem. Further research is planned to study how early can these diagnostic biomarkers predict an onset of brain states.

Prediction of the tensile strength of a tropical wood species Terminalia superba assembled by gluing: a comparative intelligent study

ABSTRACT The present study deals with the prediction of the strength of the glue joint stressed in tension on a local wood species called Terminalia Superba (Fraké) assembled according to the bevel configuration using two artificial intelligence models namely ANFIS and LSTM. The experimental data obtained during tensile tests on a tropical species allowed us to determine the mechanical properties taken as structural parameters for the LSTM and ANFIS models. The results of the analysis show that among all the LSTM methods, LSTM ‘ADAM’ offers a low root mean square error (RMSE), a high accuracy (Acc) (RMSE = 2.16, Acc = 0.756). For all methods, ANFIS obtained the best results, a high R-squared and a very low root mean square error (RMSE) (R-squared = 0.979, RMSE = 0.51). This indicates that the prediction of the tensile strength of the adhesive joint is more satisfactory with ANFIS than with LSTM.

Метод интеллектуального анализа текстовой информации для психиатрической диагностики

The automated depression detection system is a progressive technique in terms of improving clinical diagnosis and early medical intervention in cases where depression can have the most serious consequences, including self-harm or suicide. An innovative method of automated detection of depression based on textual data of patients is proposed. The developed method includes modern technologies such as the architecture of the recurrent neural network LSTM and various methods of text vectorization. Experiments conducted on publicly available datasets have confirmed the high efficiency and accuracy of the proposed method compared to the approaches used today. A unique feature of the method is the use of textual characteristics, which ensures the safety of the data provided by patients and eliminates their distortion. This approach not only increases the reliability of the results, but also avoids potential distortion of information in the analysis process. The developed method of automatic assessment of depression has high accuracy and does not require the presence of a doctor, which significantly increases the effectiveness of the process of identifying and assessing the level of depression. This approach can become a promising direction in the development of automated mental health support systems, reducing reaction time and providing more prompt assistance. In the future, the research will include training the model on data in Russian and further tuning of methods, as well as expanding the use of GloVe vectorization to improve contextual understanding of textual data. These steps are aimed at creating a more adapted and effective system for detecting depression in various linguistic contexts.

Four-Dimensional Aircraft Trajectory Prediction Based on Generative Deep Learning

Aircraft four-dimensional (4D) trajectory prediction (TP) is among the critical techniques of current existing automation systems in air traffic management (ATM), which is also the basis of future airspace operation models. This paper proposes a 4D TP method based on a conditional tabular generative adversarial network (CTGAN). The model is trained and tested using historical 4D trajectory data from the Hong Kong region. It is then compared with the most extensively researched conditional generative adversarial network (CGAN) and LSTM methods in the field of TP. The results show that Jensen–Shannon (JS) divergence, maximum mean discrepancy (MMD), and mean Euclidean distance (MED) of the model evaluation metrics for the selected route are 0.0539, 0.0150, and 0.0085, respectively. This confirms that the model for predicting 4D trajectory can accurately learn complex airspace’s trajectory distribution characteristics. Compared with the CTGAN algorithms shown in previous studies, the enhanced CTGAN approach developed in our research can reduce around 10% running time and 20% MED. Compared to CGAN, it results in an 81.55% reduction in MED. Compared to LSTM, when predicting with a 10-minute time span, the average mean absolute error for parameters latitude (Lat), longitude (Lon), geometric height (GH), and ground speed (GS) decreases by 7.76, 84.48, 81.98, and 36.51%, respectively.

Revisiting the potential value of vital signs in the real-time prediction of mortality risk in intensive care unit patients

BackgroundPredicting patient mortality risk facilitates early intervention in intensive care unit (ICU) patients at greater risk of disease progression. This study applies machine learning methods to multidimensional clinical data to dynamically predict mortality risk in ICU patients.MethodsA total of 33,798 patients in the MIMIC-III database were collected. An integrated model NIMRF (Network Integrating Memory Module and Random Forest) based on multidimensional variables such as vital sign variables and laboratory variables was developed to predict the risk of death for ICU patients in four non overlapping time windows of 0–1 h, 1–3 h, 3–6 h, and 6–12 h. Mortality risk in four nonoverlapping time windows of 12 h was externally validated on data from 889 patients in the respiratory critical care unit of the Chinese PLA General Hospital and compared with LSTM, random forest and time-dependent cox regression model (survival analysis) methods. We also interpret the developed model to obtain important factors for predicting mortality risk across time windows. The code can be found in https://github.com/wyuexiao/NIMRF.ResultsThe NIMRF model developed in this study could predict the risk of death in four nonoverlapping time windows (0–1 h, 1–3 h, 3–6 h, 6–12 h) after any time point in ICU patients, and in internal data validation, it is suggested that the model is more accurate than LSTM, random forest prediction and time-dependent cox regression model (area under receiver operating characteristic curve, or AUC, 0–1 h: 0.8015 [95% CI 0.7725–0.8304] vs. 0.7144 [95%] CI 0.6824–0.7464] vs. 0.7606 [95% CI 0.7300–0.7913] vs 0.3867 [95% CI 0.3573–0.4161]; 1–3 h: 0.7100 [95% CI 0.6777–0.7423] vs. 0.6389 [95% CI 0.6055–0.6723] vs. 0.6992 [95% CI 0.6667–0.7318] vs 0.3854 [95% CI 0.3559–0.4150]; 3–6 h: 0.6760 [95% CI 0.6425–0.7097] vs. 0.5964 [95% CI 0.5622–0.6306] vs. 0.6760 [95% CI 0.6427–0.7099] vs 0.3967 [95% CI 0.3662–0.4271]; 6–12 h: 0.6380 [0.6031–0.6729] vs. 0.6032 [0.5705–0.6406] vs. 0.6055 [0.5682–0.6383] vs 0.4023 [95% CI 0.3709–0.4337]). External validation was performed on the data of patients in the respiratory critical care unit of the Chinese PLA General Hospital. Compared with LSTM, random forest and time-dependent cox regression model, the NIMRF model was still the best, with an AUC of 0.9366 [95% CI 0.9157–0.9575 for predicting death risk in 0–1 h]. The corresponding AUCs of LSTM, random forest and time-dependent cox regression model were 0.9263 [95% CI 0.9039–0.9486], 0.7437 [95% CI 0.7083–0.7791] and 0.2447 [95% CI 0.2202–0.2692], respectively. Interpretation of the model revealed that vital signs (systolic blood pressure, heart rate, diastolic blood pressure, respiratory rate, and body temperature) were highly correlated with events of death.ConclusionUsing the NIMRF model can integrate ICU multidimensional variable data, especially vital sign variable data, to accurately predict the death events of ICU patients. These predictions can assist clinicians in choosing more timely and precise treatment methods and interventions and, more importantly, can reduce invasive procedures and save medical costs.

The Prediction of the Remaining Useful Life of Rotating Machinery Based on an Adaptive Maximum Second-Order Cyclostationarity Blind Deconvolution and a Convolutional LSTM Autoencoder.

The prediction of the remaining useful life (RUL) is important for the conditions of rotating machinery to maintain reliability and decrease losses. This study proposes an efficient approach based on an adaptive maximum second-order cyclostationarity blind deconvolution (ACYCBD) and a convolutional LSTM autoencoder to achieve the feature extraction, health index analysis, and RUL prediction for rotating machinery. First, the ACYCBD is used to filter noise from the vibration signals. Second, based on the peak value properties, a novel health index (HI) is designed to analyze the health conditions for the denoising signal, showing a high sensitivity for the degradation of bearings. Finally, for better prognostics and health management of the rotating machinery, based on convolutional layers and LSTM, an autoencoder can achieve a transform convolutional LSTM network to develop a convolutional LSTM autoencoder (ALSTM) model that can be applied to forecast the health trend for rotating machinery. Compared with the SVM, CNN, LSTM, GRU, and DTGRU methods, our experiments demonstrate that the proposed approach has the greatest performance for the prediction of the remaining useful life of rotating machinery.

Ensemble Learning-based Algorithms for Traffic Flow Prediction in Smart Traffic Systems

Due to the tremendous growth of road traffic accidents, Intelligent Transportation Systems (ITSs) are becoming even more important. To prevent road traffic accidents in the long term, it is necessary to find new vehicle flow management techniques in order to optimize traffic flow. With the high growth of deep learning and machine learning, these methods are increasingly being used in ITSs. This research provides a novel conceptual ITS model that aims to predict vehicle movement through the collective learning usage to anticipate intersections. The proposed approach consists of three main stages: data collection through cameras and sensors, implementation of machine learning and deep learning algorithms, and result evaluation, utilizing the coefficient of determination (R-squared), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). To accomplish this, various machine learning and deep learning algorithms, such as Random Forest, LSTM, Linear Regression, and ensemble methods (bagging), were incorporated into the model. The findings revealed the enhancement due to the proposed method, which was observed through a significant performance improvement of 93.52%.

Ultra-short term wind power prediction based on quadratic variational mode decomposition and multi-model fusion of deep learning

To address the challenges associated with the intricate selection of decomposition technical parameters and the adverse impact of high-frequency non-stationary components on prediction accuracy, a differential evolution (DE) and sparrow search algorithm (SSA) parameter optimization model is formulated. Through the optimization of variational mode decomposition (VMD) and bidirectional short and long-term memory (BiLSTM) neural networks, enhancements are made to the overall performance of the prediction method. Subsequently, DESSA–VMD is employed to quadratically decompose the high-frequency unstable component, mitigating the detrimental effects of randomness and volatility on the prediction accuracy of the primary component. Following this, the DESSA–BiLSTM neural network is applied to predict the high-frequency strong non-stationary component, whereas the autoregressive integrated moving average is utilized for predicting the low-frequency periodic component. Finally, a case analysis is conducted using actual data comprising 2000 sets of wind power in a province. When the test set consisted of 600 groups, the root mean square error index value of the proposed prediction model exhibited reductions of 1.95 MW, 2.86 MW, and 1.16 MW, respectively, compared with the BP, SVM, and LSTM methods. Additionally, the mean absolute percentage error index showed decreases of 2.26 %, 2.96 %, and 1.73 %, respectively.

Novel cloud-AIoT fault diagnosis for industrial diesel generators based hybrid deep learning CNN-BGRU algorithm

The industrial diesel generator is essential to provide electric power during emergencies and requires accurate failure diagnosis in preventative maintenance services. This research presents a novel AIoT system-based hybrid deep learning algorithm for online monitoring of the anomaly conditions of 125 kW/250 kW industrial generators. Firstly, the IoT modules with different precise sensors are developed to be installed on 125 kW and 250 kW industrial generators to simulate other faults conditions in the laboratory. Then, the supervised learning operations are deployed to train deep learning algorithms with collected labeled data. The proposed algorithm, a hybrid convolution neural network bidirectional gate recurrent unit (CNN-BGRU), is deployed to extract deep features from long sequential historical signals and classify the anomaly conditions. The experiments in different 10 s, 20 s, and 30 s sampling frequencies are performed to evaluate and analyze with various benchmarks, which demonstrate efficiencies of the proposed CNN-BGRU method with other traditional deep learning algorithms, including RNN, CNN, GRU, LSTM, BRNN, and BGRU methods. The proposed hybrid method accomplishes the supreme augmentation of 29.43 % loss, 36.55 % validating loss, 25.86 % MAE, 27.00 % validating MAE, 16.78 % MSE, 20.64 % validating MSE, 6.36 % Pre, 25.34 % validating Pre, 17.30 % Rec, and 26.91 % validating Rec compared with state-of-the-art approaches. The AIoT system-based hybrid CNN-BGRU algorithm provides better improvement and higher accuracy in fault diagnosis of firefighting pumps for maintenance service in Industry 4.0.