Minimum Mean Absolute Error Research Articles

A New Algorithm for Estimating Low Cloud-Base Height in Southwest China

Abstract The prevalence of low clouds significantly affects flight safety in Southwest China. However, relevant cloud parameters, especially low cloud-base height (LCBH), lack accurate forecasts. Based on the hourly atmospheric vertical profiles of ERA5 from 2008 to 2019, we developed a new algorithm for estimating LCBH by combining relative humidity (RH) threshold methods with convective condensation level (CCL) (RHs-CCL). To evaluate the performance of RHs-CCL, we use it to estimate the hourly LCBH of airports in Southwest China and compare the results with those based on the ground-based observations and the ERA5 CBH data. Using the observations as a ground truth, we compare the RHs-CCL algorithm with several existing algorithms with the following findings: 1) The correlation coefficient between RHs-CCL and observations reaches 0.5 on average, and the error of RHs-CCL is smaller than those of existing algorithms, with the minimum mean absolute error and root-mean-square error at the four airports studies being able to reach 243 and 321 m. 2) The bias score of RHs-CCL is 0.97 on average, and low clouds classification utilizing RHs-CCL attains the highest accuracy, up to 86%. 3) The errors of ERA5 CBH are the largest when compared with the others. 4) By implementing convective cloud occurrence condition and CCL, RHs-CCL has better applicability in regions of enhanced convective activity. These results suggest the potential of RHs-CCL as an algorithm moving forward for improvement of the LCBH estimates based upon high-resolution reanalysis products and for better predictions of the LCBH utilizing outputs from numerical weather prediction models. Significance Statement The new algorithm developed in this study can accurately estimate low cloud-base heights from vertical profiles of atmospheric variables. It provides us a much more computationally efficient approach for predicting low cloud-base height relative to running cloud models, which is critical for weather forecasting at locations lacking computational resources and/or cloud modeling capability. In areas such as Southwest China, low clouds are very common, and they pose major threats to aviation safety. The new algorithm has been successfully integrated into the daily operation at Guiyang Airport in Southwest China and demonstrated excellent skills in estimating cloud-base heights. The implementation of the algorithm in aviation forecasting over a broader region is on the horizon.

Prognostics and health management for piezoresistive pressure sensor based on improved gated recurrent unit networks

Application fields for piezoresistive pressure sensors have become increasingly extensive in recent years. So a high reliability of the sensor is also required. However, considering that some sensors operate in hostile environments,the need to ensure continuous operation accuracy, prognostics, and health management (PHM) for piezoresistive pressure sensors should not be ignored. To solve this problem, a fault diagnosis and prognostic method that combines a support vector machine (SVM) and deep gated recurrent unit (DGRU) network optimized by hunter–prey optimization (HPO) is proposed in this paper. First, three sensor fault types are defined. Second, SVM is adopted to realize the fault diagnosis. Third, two layers of DGRU are employed to predict the health index, which is defined to represent the health state of the sensor. Meanwhile, the optimal parameters of the DGRU are optimized by HPO algorithms. Finally, the remaining useful life can be estimated by the predicted health index and failure threshold. The method proposed in this paper is proved to be effective and accurate. The fault diagnosis accuracy is 100% in the three fault types defined by this paper. The minimum mean absolute error is 6 × 10−5. It proves the proposed method of PHM in this paper is viable in a real application.

Groundwater level prediction with machine learning for the Vidisha district, a semi-arid region of Central India

Groundwater is one of the most important freshwater resources, particularly in dry and semi-arid regions with low annual precipitation and frequent droughts. For the planning and the management of water resources, this is very essential to generate a soft computing method that can predict or forecast groundwater table (GWT). The different machine and deep learning algorithms are evaluated in this research work to predict the groundwater table using season, monthly rainfall, and evapotranspiration rate, which are further compared. The performance evaluation of the different algorithms was done using five different statistical parameters, i.e., mean absolute error (MAE), mean squared error (MSE), root-mean-square error (RMSE), R2−score, and Nash-Sutcliffe Efficiency (NSE). From the results of this study, it is concluded that the Multivariate polynomial regression algorithm is the most efficient for predicting the groundwater table with the minimum mean absolute error in the range of 0.63–0.85 and R2-score of 0.82–0.88.

Process monitoring in submerged abrasive waterjet cutting of Ti6Al4V by vibration signal

ABSTRACTThe submerged AWJ reduces the supersonic cutting noise, reduces the divergence of the jet, and improves the performance characteristics, but affects the visibility of the workpiece. Hence, this work deals with the monitoring of the combined effect of responses, termed as the multi-response performance index (MRPI), by vibration signals. Firstly, the unsubmerged and submerged AWJs are compared, and the submerged condition gave an average improvement of 1.9 mm in depth of cut, 2.1 mm in smooth cutting zone, 1.03 mm in kerf width, and 1.4 µm in roughness. Furthermore, the TOPSIS method is applied to combine the aforementioned responses into MRPI. The wavelet transform (WT) was used to extract information from the vibration signals, and the features selected by principal component analysis were fed as inputs to ANN and ANFIS techniques. In ANN, 20 architectures were compared for the prediction of MRPI, and the architecture 4-8-4-1 gave a minimum mean absolute error (MAE) of 0.065, and ANFIS gave a minimum MAE of 0.03. Overall, the integrated WT-ANFIS efficiently predicted the MRPI. Finally, the most influential parameters affecting the MRPI were found to be the pressure followed by the transverse rate with contributions of 44% and 21%.

Heterogeneous Transfer Learning for Wi-Fi Indoor Positioning Based Hybrid Feature Selection.

This paper presents the application of heterogeneous transfer learning (HetTL) methods which consider hybrid feature selection to reduce the training calibration effort and the noise generated by fingerprint duplicates obtained from multiple Wi-Fi access points. The Cramer–Rao Lower Bound analysis (CRLB) was also applied to evaluate and estimate a lower limit for the variance of a parameter estimator used to analyze positioning performance. We developed two novel algorithms for feature selection in fingerprint-based indoor positioning problems (IPP) to enhance positioning performance in the target domain with the HetTL. The algorithms comprised two scenarios: (i) a principal component analysis-based approach (PCA-based) and (ii) a hybrid approach that takes both PCA and correlation effect analysis into account (hybrid scenario). Accordingly, a new feature vector was constructed by retaining only the most significant predictors, and the most efficient feature dimensions were also determined by using a hybrid-based approach. Experimental results showed that the hybrid-based proposed algorithm has the minimum mean absolute error. The CRLB analysis also showed that the number of Wi-Fi access points could affect the lower bound location estimation error; however, identifying the most significant predictors is an effective approach to improve positioning performance.

Predicting Crude Oil Price in Nigeria with Machine Learning Models

Background: The rise in crude oil prices yields serious consequences for both oil-producing and non-oil-producing countries. The increase in global commodity prices contributes to the financial income and foreign exchange reserves of oil-exporting countries. However, countries such as Nigeria that sell crude oil and purchase refined fuel confront more complicated situations. To this end, there exists a need to obtain a robust prediction model for the crude oil price of Nigeria. Objective: This study is to determine the best model among the machine-learning time series models considered to predict crude oil prices in Nigeria. Methods: The alternative models were the auto-regressive integrated moving average model, Naive Bayes, Holtwinter trend model, exponential smoothing model, and neural network autoregressive (NNETAR) model. The prediction criteria adopted for model screening were the root mean square error (RMSE), mean absolute error (MAE) and mean absolute percentage error (MAPE). Daily crude oil prices in dollars obtained from the Central Bank of Nigeria were used for analysis spanning from October 1, 2009 to March 22, 2022 with 2836 data points. Results: The NNETAR model showed the minimum RMSE, MAE, and MAPE for cross-validation sets considered. Conclusion: The NNETAR model was recommended for the prediction of crude oil prices in Nigeria.

Mammalian eye gene expression using support vector regression to evaluate a strategy for detecting human eye disease

Background and purpose: Machine learning is a class of modern and strong tools that can solve many important problems that nowadays Humans may be faced with. Support Vector Regression (SVR) is a way to build a regression model which is an incredible member of the machine learning family. SVR has been proven to be an effective tool in real-value function estimation. As a supervised-learning approach, SVR trains using a symmetrical loss function, which equally penalizes high and low misestimates. Recently, high-dimensional datasets are the most challenging problem that may be faced. The main problems in high-dimensional data are the estimation of the coefficients and interpretation. In the high-dimension problems, classical methods are not applicable because of a large number of predictor variables. SVR is an excellent alternative method to analyze such datasets. One of the main advantages of SVR is that its computational complexity does not depend on the dimensionality of the input space. Additionally, it has excellent generalization capability, with high prediction accuracy.
 Methods: SVR is one of the best methods to analyze high-dimensional datasets. It is a really reliable and robust approach to have a good fit with high accuracy. SVR uses the same principles as the support vector machine for classification, with only a few minor differences.
 Results: The techniques for analyzing the high-dimension datasets are really important methods because we frequently face such datasets in medical science and gene expression. It is not easy to analyze the high-dimension datasets because the classic methods cannot be used to estimate and interpret them. Therefore, we have to use alternative methods to analyze them. SVR is one of the best methods that can be applied. In this research, SVR is used in a real high-dimension dataset about the gene expression in eye disease, and then it is compared with well-known methods: LASSO and Sparse least trimmed squared (sparse LTS) methods. Based on the numerical result, SVR and Sparse LTS were better than LASSO, since the real dataset contained outliers (bad observation with big residuals).
 Conclusions: SVR method was the best method to model and predict the high-dimensional mammalian eye dataset, because it was not affected by the outliers' corruptive impact, and it has minimum MSE (mean squares error), MAE (mean absolute error) and RMSE (root mean squared error) fitting criteria in comparison with the classical methods such as LASSO and sparse LTS estimations. Thus, sparse LTS was found to act better than the LASSO method. Moreover, stabilization of the data and freedom from obtaining the regularization parameter by running a complicated algorithmic program, which decreased the computational costs dramatically, were the invaluable advantages of this technique in comparison with the classical methods.

Neural network estimations of annealed and non-annealed Schottky diode characteristics at wide temperatures range

In this study, Artificial Neural Network (ANN) model has been proposed to characterize the annealed and the non-annealed Schottky diode from experimental data. The experimental current values of Ni/n-type 6H–SiC Schottky diode for the voltages applied to the diode terminal starting from 80 K with 20 K steps up to 500 K temperature were measured for both non-annealed and annealed Schottky diodes. The applied voltage has been varied starting from -2 V with 10 mV steps up to +2 V for each temperature value. The modeling performance has been assessed according to the varying number of neurons in the hidden layer, starting from 5 to 50 neurons, thereafter the optimum number of neurons has been obtained for both annealed and non-annealed ANN models. The minimum Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) indices values for both annealed and non-annealed diodes have been obtained with 40 neurons for both the training and test phase.

Prediction of landslide tsunami run-up on a plane beach through feature selected MLP-based model

We proposed new prediction models based on multilayer perceptron (MLP) which successfully predict the maximum run-up of landslide-generated tsunami waves and assess the role of parameters affecting it. The input is approximately 55,000 rows of data generated through an analytical solution employing slide’s cross section, initial submergence, vertical thickness, horizontal length, beach slope angle and the maximum run-up itself, along with its occurrence time. The parameters are first ranked through a feature selection algorithm and six models are constructed for a 9,000-row randomly sampled dataset. These MLP-based models led predictions with a minimum Mean Absolute Percentage Error of 1.1% and revealed that vertical slide thickness has the largest impact on the maximum tsunami run-up, whereas beach slope angle has minimal effect. Comparison with existing literature showed the reliability and applicability of the offered models. The methodology introduced here can be suggested as fast and flexible method for prediction of landslide-induced tsunami run-up.

Multitasking Learning Model Based on Hierarchical Attention Network for Arabic Sentiment Analysis Classification

Limited approaches have been applied to Arabic sentiment analysis for a five-point classification problem. These approaches are based on single task learning with a handcrafted feature, which does not provide robust sentence representation. Recently, hierarchical attention networks have performed outstandingly well. However, when training such models as single-task learning, these models do not exhibit superior performance and robust latent feature representation in the case of a small amount of data, specifically on the Arabic language, which is considered a low-resource language. Moreover, these models are based on single task learning and do not consider the related tasks, such as ternary and binary tasks (cross-task transfer). Centered on these shortcomings, we regard five ternary tasks as relative. We propose a multitask learning model based on hierarchical attention network (MTLHAN) to learn the best sentence representation and model generalization, with shared word encoder and attention network across both tasks, by training three-polarity and five-polarity Arabic sentiment analysis tasks alternately and jointly. Experimental results showed outstanding performance of the proposed model, with high accuracy of 83.98%, 87.68%, and 84.59 on LABR, HARD, and BRAD datasets, respectively, and a minimum macro mean absolute error of 0.632% on the Arabic tweets dataset for five-point Arabic sentiment classification problem.

A CNN model for predicting size of buried objects from GPR B-Scans

A convolutional neural networks (CNN) model for predicting size of buried objects from ground penetrating radar (GPR) B-Scans is proposed. As a pre-processing step, Sobel, Laplacian, Scharr, and Canny operators are used for edge detection of the hyperbolic features. The proposed CNN architecture extracts high level signatures in the initial stages of the model and learns additional low-level features when the input data passes through the neural network to finally make an estimation of the required parameter. Artificially generated GPR B-Scans are used to train the model. The proposed method demonstrates good performance in predicting buried object size. Upon comparison, Scharr operator followed by a deep CNN model showed the best performance, having the minimum mean absolute percentage error of 6.74 when tested on new, unseen data.

Improvement towards Prediction Accuracy of Principle Mineral Resources Using Threshold

The production data of mineral resources are noisy, nonstationary, and nonlinear. Therefore, some techniques are required to address the problem of nonstationarity and complexity of noises in it. In this paper, two hybrid models (EMD-CEEMDAN-EBT-MM and WA-CEEMDAN-EBT-MM) flourish to improve mineral production prediction. First, we use empirical mode decomposition (EMD) and wavelet analysis (WA) to denoise the data. Second, ensemble empirical mode decomposition (EEMD) and complete ensemble empirical mode decomposition (CEEMDAN) are used for the decomposition of nonstationary data into intrinsic mode function (IMF). Then, empirical Bayesian threshold (EBT) is applied on noise dominant IMFs to consolidate noises, which are further used as input in the data-driven model. Next, other noise-free IMFs are used in the stochastic model as input for the prediction of minerals. At last, the predicted IMFs are ensemble for final prediction. The proposed strategy is exemplified using Pakistan's four major mineral resources. To measure the prediction performance of all the models, three methods, that is, mean relative error, mean square error, and mean absolute percentage error, are used. Our proposed framework WA-CEEMDAN-EBT-MM has shown improvement with minimum mean absolute percentage error value compared to other existing models in prediction accuracy for all four minerals. Therefore, our proposed strategy can predict the noisy and nonstationary time-series data with an efficient mechanism. Hence, it will be helpful to the policymakers for making policies and planning in mineral resource management.

A weakly-supervised approach for flower/fruit counting in apple orchards

Flower and fruit count is a critical metric in developing crop-load management and harvesting strategies during flower/fruit development and harvest seasons. Growers currently rely on their prior experience and/or manual count in sample areas/trees to estimate the number of flowers/fruits in orchards. In this work, we propose a simplified yet robust deep learning-based weakly-supervised flower/fruit Counting Network (CountNet) and investigate its accuracy in commercial orchard images. Unlike detection-based counting methods, which require individual object detection, CountNet learns from image-level annotation with the number of objects (flowers or fruits) as input without explicitly specifying the object’s signature and location. Experiments were conducted in images acquired in an unstructured commercial orchard environment. Results showed a minimum Mean Absolute Error (MAE)/Root Mean Square Error (RMSE) of 12.0/18.4 and 2.9/4.3 for the apple flower and fruit dataset respectively. Activated region/feature visualization techniques revealed that CountNet is looking into different apple flower/fruit edges and features to make the count decisions. The results are promising in simplifying the automated methods for flower/fruit counting which can lead to reduced manual counting in the field, manual image annotation, and computational complexity and memory requirement of the object counting system.

A modified one parameter Liu estimator for Conway-Maxwell Poisson response model

The maximum likelihood estimator (MLE) is generally used to estimate the Conway Maxwell Poisson regression model (COMPRM). When the explanatory variables are highly correlated, then the MLE results are not valid. In this study, we proposed a modified one-parameter Liu estimator in the presence of multicollinearity among the regressors for the COMPRM. The theoretical properties of the proposed estimator are derived and compared it with the available biased estimators as well as the MLE based on the matrix mean squared error (MSE) and scalar MSE criteria. To investigate the efficiency of the proposed estimator, a Monte Carlo simulation analysis was performed under various controlled conditions. Finally, two real applications are considered in the superiority of the proposed estimator. The simulation and real applications results show that the proposed estimator outperforms the classical MLE and other biased estimators in terms of the minimum MSE and mean absolute error criterion .

Application of ensemble machine learning methods for kerogen type estimation from petrophysical well logs

The current study is the first report of estimating kerogen type from petrophysical well logs implementing various machine learning techniques. The methodology is explained through a case study from the Perth Basin, Western Australia. Firstly, the statistical relationships between the petrophysical data (including gamma ray (GR), sonic (DTCO), neutron (NEUT) and density (RHOB) logs) and the Rock-Eval derived hydrogen index (HI) and oxygen index (OI) are investigated. Afterwards, the various machine learning (ML) techniques, including radial basis function (RBF) and multi-layer perceptron (MLP) artificial neural networks, random forest (RF), support vector machine (SVM) and decision tree (DT), are applied to estimate the hydrogen and oxygen index values. Additionally, the MLP network is optimized by the grey wolf optimization (GWO), genetic algorithm (GA) and particle swarm optimization (PSO). The outputs of the various ML approaches are integrated, employing both simple averaging and weighted averaging committee machines. Three statistical parameters of root mean square error (RMSE), mean absolute error (MAE) and correlation coefficient (R) are employed to evaluate and compare the performance of the machine learning ensembles. In general, employing optimization algorithms on the MLP network reinforces the network's performance for the hydrogen index (HI) estimation by increasing overall R and decreasing RMSE and MAE for test data. Likewise, RMSE and MAE values of test data for the oxygen index (OI) decrease using all three optimizers, although R values decrease as well. However, GWO is the most efficient optimizer in diminishing RMSE and MAE and rising R values of test data for the hydrogen index estimation. This algorithm also provides the minimum RMSE and MAE for the oxygen index (OI) estimation. Ultimately, among the proposed intelligent approaches, the weighted averaging committee machine (WACM) provides the maximum correlation coefficient (R) and minimum errors (RMSE and MAE) for both hydrogen index (HI) and oxygen index (OI) estimating. The estimated oxygen and hydrogen indices values are then successfully employed to predict the kerogen types based on the van Krevelen diagram.

Deep Learning Based Soft Sensor to Predict Total Suspended Solids of Refinery Water Treatment Plant Using Real Process Data

Soft sensors have proven themselves as a valuable tool in terms of quality prediction, process monitoring, process control and minimizing operating cost. Compared to traditional hardware sensors, a soft sensor is a technology that consists of all combinations of mathematical model, data processing and software techniques. Deep learning method has the great potential to be data-driven soft sensors. In this paper, Gated Recurrent Unit (GRU) which is one of the deep learning model was developed as soft sensors for prediction of Total Suspended Solids (TSS) concentration having a role of water quality of New Water Treatment Plant (NWTP) in oil refinery. A real dataset consisting of 39 parameters (41,123 observations) over 6 months from different sensor measurements including flow, temperature, pressure and turbidity etc. was used for GRU model. Also, Principal Component Analysis (PCA) was used for reducing the dimensionality of datasets, providing data interpretation and examining data that correlates with each other. In this case, three different scenarios were created using 5, 10 and 39 process variables as input based on the PCA results. The results show that acceptable estimation has been achieved through the developed algorithm. The minimum mean absolute errors of 0.0108 % for testing dataset and R2 of 0.899 for using 10 input variables that affect TSS concentration were achieved. The motivation in this study is that the data-driven predictive model is realized from historical data in large scale system, so it is valuable for the theory to take shape in an industrial application.

Time Series Data Error Optimization through Three Stacked Long Short-Term Memory

Evidence gathered on the same component over a period at various points in time are known as time series. In order to make predictions and guide strategic decision-making, time series forecasting means examining time series da-ta using statistics and simulation. Due to the various components in time series data, prediction of any target element have high risk of failure and still remains a challenge for the researchers. This paper propose three stacked long short-term memory that utilize the significant huge Stock Market Dataset extracted from the KAGGLE database. The Stock Market dataset contains 1344 exchange traded fund data and 7195 stock data. The data is pre-processed with encoding and incomplete values. The prescriptive information analysis have been accomplished to investigate the time series distribution. The original dataset applied to one, two, three and four stacked long short-term memory to analyze the error performance stock rate prediction. The dataset also applied to one, two, three and four stacked long short-term memory with dropout to analyze the error performance stock rate prediction. The performance metrics utilized for analysis are Mean Squared Error, Mean Absolute Error and Root Mean Squared error. Implementation results exhibits that proposed three-stacked long short-term memory shows minimum Mean Absolute Error of 0.004384, Mean Squared Error of 0.000045 and Root Mean Squared error of 0.006680 when compared with all other long short-term memory models.

Prediction of Liver Weight Recovery by an Integrated Metabolomics and Machine Learning Approach After 2/3 Partial Hepatectomy.

Liver has an ability to regenerate itself in mammals, whereas the mechanism has not been fully explained. Here we used a GC/MS-based metabolomic method to profile the dynamic endogenous metabolic change in the serum of C57BL/6J mice at different times after 2/3 partial hepatectomy (PHx), and nine machine learning methods including Least Absolute Shrinkage and Selection Operator Regression (LASSO), Partial Least Squares Regression (PLS), Principal Components Regression (PCR), k-Nearest Neighbors (KNN), Support Vector Machines (SVM), Random Forest (RF), eXtreme Gradient Boosting (xgbDART), Neural Network (NNET) and Bayesian Regularized Neural Network (BRNN) were used for regression between the liver index and metabolomic data at different stages of liver regeneration. We found a tree-based random forest method that had the minimum average Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and the maximum R square (R2) and is time-saving. Furthermore, variable of importance in the project (VIP) analysis of RF method was performed and metabolites with VIP ranked top 20 were selected as the most critical metabolites contributing to the model. Ornithine, phenylalanine, 2-hydroxybutyric acid, lysine, etc. were chosen as the most important metabolites which had strong correlations with the liver index. Further pathway analysis found Arginine biosynthesis, Pantothenate and CoA biosynthesis, Galactose metabolism, Valine, leucine and isoleucine degradation were the most influenced pathways. In summary, several amino acid metabolic pathways and glucose metabolism pathway were dynamically changed during liver regeneration. The RF method showed advantages for predicting the liver index after PHx over other machine learning methods used and a metabolic clock containing four metabolites is established to predict the liver index during liver regeneration.

Design of filter coefficients for optimal factored truncated cascade FIR filter using optimization algorithm

Signal filtering acts as one of the basic requirement of communication networks for the removal of unwanted features from the signal. The design of appropriate digital filter requires the selection of optimal filter coefficients for the generation of desired frequency response with reduced hardware complexity. This paper proposes a hybrid optimization algorithm named as Brain Storm- Grey Wolf Optimizer (BSGWO) algorithm for the selection of filter coefficients in the design of factored truncated cascade FIR filter. The proposed algorithm is the hybridization of the optimization algorithms, namely Brain Storm Optimization (BSO) and Grey Wolf Optimizer (GWO). The input signal is interpolated initially for the formation of an intermediate signal using the FIR filter. Then, the factored truncated cascade filter is developed for the interpolation of the signal. After designing the filter coefficients, the optimal selection of the filter coefficients is performed using the proposed BSGWO algorithm. The original filter is developed with the use of the least square estimation and the new filter is developed using the proposed algorithm that tunes the filter coefficients. The performance of the proposed system is analyzed using the metrics, such as fitness, Mean Absolute Error (MAE), magnitude, and the number of components. The proposed method produces minimum fitness, MAE, magnitude and number of components of 0.05, 0.0155, − 96.0 dB and 3372, respectively that shows the effectiveness of the proposed method.

An efficient approach for detecting anomalous events in real‐time weather datasets

AbstractEvent detection in real‐time is applied in diverse domains such as detection of fraudulent activities in commercial transactions, detection of faulty systems in industries, and so forth. Businesses and organizations benefit from the actionable information obtained through various techniques available for anomalous event detection. Real‐time event detection is nowadays handled through streaming data frameworks. Traditional approaches effectively handle event detection in real‐time but with more false positives, thus, resulting in false alarms. In this article, an efficient approach comprising two components, an offline model and an online event detection pipeline, is proposed to achieve minimum mean absolute error (MAE). An offline module is developed to investigate a variety of deep learning models that prove suitable for event detection in real‐time. The experiments conducted with PubNub sensors datasets demonstrate that the long short‐term memory unit of recurrent neural networks is the best suitable model for anomalous event detection. The online pipeline module is built using streaming data frameworks to predict the abnormal peaks. It is revealed through the experimental results that the proposed approach efficiently detects anomalous events in real‐time and also eliminates false positives.