The Long Short-Term Memory Algorithm and the Autoregressive Integrated Moving Average Approach in Business Tendency Survey

Forecasting peak electrical energy consumption is important because it allows utilities to properly plan for the production and distribution of electrical energy. This reduces operating costs and avoids power outages. In addition, it can help reduce environmental impact by allowing for more efficient power generation and reducing the need for additional fossil fuels during periods of high demand. In the current work, electric power consumption data from “Compagnie Electrique du Benin (CEB)” was used to deduce the peak electric power consumption at peak hours. The peak consumption of electric power was predicted using hybrid approaches based on traditional time series prediction methods (autoregressive integrated moving average (ARIMA)) and deep learning methods (long short-term memory (LSTM), gated recurrent unit (GRU)). The ARIMA approach was used to model the trend term, while deep learning approaches were employed to interpret the fluctuation term, and the outputs from these models were combined to provide the final result. The hybrid approach, ARIMA-LSTM, provided the best prediction performance with root mean square error (RMSE) of 7.35, while for the ARIMA-GRU hybrid approach, the RMSE was 9.60. Overall, the hybrid approaches outperformed the single approaches, such as GRU, LSTM, and ARIMA, which exhibited RMSE values of 18.11, 18.74, and 49.90, respectively.

Optimized long short-term memory (LSTM) network for performance prediction in unconventional reservoirs

Abstract. Flood forecasting systems play a key role in mitigating socioeconomic damage caused by flood events. The majority of these systems rely on process-based hydrologic models (PBHMs), which are used to predict future runoff. Many operational flood forecasting systems additionally implement models aimed at enhancing the predictions of the PBHM, either by updating the PBHM's state variables in real time or by enhancing its forecasts in a post-processing step. For the latter, autoregressive integrated moving average (ARIMA) models are frequently employed. Despite their high popularity in flood forecasting, studies have pointed out potential shortcomings of ARIMA-type models, such as a decline in forecast accuracy with increasing lead time. In this study, we investigate the potential of long short-term memory (LSTM) networks for enhancing the forecast accuracy of an underperforming PBHM and evaluate whether they are able to overcome some of the challenges presented by ARIMA models. To achieve this, we developed two hindcast–forecast LSTM models and compared their forecast accuracies to that of a more conventional ARIMA model. To ensure comparability, one LSTM was restricted to use the same data as ARIMA (eLSTM), namely observed and simulated discharge, while the other additionally incorporated meteorologic forcings (PBHM-HLSTM). Considering the PBHM's poor performance, we further evaluated if the PBHM-HLSTM was able to extract valuable information from the PBHM's results by analyzing the relative importance of each input feature. Contrary to ARIMA, the LSTM networks were able to mostly sustain a high forecast accuracy for longer lead times. Furthermore, the PBHM-HLSTM also achieved a high prediction accuracy for flood events, which was not the case for ARIMA or the eLSTM. Our results also revealed that the PBHM-HLSTM relied, to some degree, on the PBHM's results, despite its mostly poor performance. Our results suggest that LSTM models, especially when provided with meteorologic forcings, offer a promising alternative to frequently employed ARIMA models in operational flood forecasting systems.

To get a handle on global climate patterns, we need to accurately forecast the El Niño Modoki Index (EMI). This study compares two methods for predicting the EMI's short-term behaviour based solely on its historical data. We pitted a classic statistical method, the AutoRegressive Integrated Moving Average (ARIMA) model, against a modern deep learning approach using a Long Short-Term Memory (LSTM) network. We trained both models on monthly EMI data from 1982 to 2022 and then tested how well they could predict the period from 2023 to 2025.Our results showed that the ARIMA(2,0,0) model works as a solid, understandable baseline, capturing the main movements of the index with a Root Mean Squared Error (RMSE) of 0.4338 and an R-squared (R²) of only 0.0533. The LSTM network, however, was much better at handling the quirky, non-linear nature of the data, leading to a far more accurate forecast with an RMSE of just 0.0820 and an R² of 0.9658. Ultimately, while a simple ARIMA model is useful as a benchmark, our work makes it clear that LSTM networks can offer a major leap forward in forecasting accuracy for complex climate indicators like the EMI.

Time series forecasting is an important topic widely addressed with traditional statistical models such as regression, and moving average. This work uses the state-of-the-art Long Short-Term Memory (LSTM) Networks to predict Ecuadorian imports of Home Appliances, and to compare the results against those obtained by traditional methods. First, an ARIMA model was used to forecast imports data. Then, the predictions were calculated by a Univariate LSTM network. The time series used in both experiments was the monthly average of imports from 1996 to April 2019. In addition, time series of GDP Growth, Population, and Inflation were included in the model to test prediction improvements. The performance of the models was assessed comparing the Mean Squared, Root Mean Square and Mean Absolute Error metrics. The results show that a LSTM network produces a better fit of the imports data and improved predictions compared against those produced by the ARIMA model. Furthermore, the use of multivariate time series (i.e., GDP Growth, Population, Inflation) data, for the LSTM model, did not produce significant improvements compared to the univariate imports time series.

For solving the problems related to prediction of time series data the Artificial Intelligence models i.e. Machine learning and deep learning are becoming popular. These models have been proven to deliver greater accuracy than traditional regression models. Among these, Recurrent Neural Networks (RNNs) with features (e.g. memory storage), such as Long Short-Term Memory (LSTM) networks, have proven to show a superior edge over models like Autoregressive Integrated Moving Average (ARIMA). LSTM networks are unique because they use special element namely "gates" which help them remember and process longer sequences of time series data. Based on hyper-parameter tuning, various LSTM model configurations are possible to be developed, each of which are designed to address specific prediction challenges and improve model performance. Thus, the key consideration is whether the elements of gates in LSTM networks alone are sufficient to deliver better predictions or if further training is necessary to enhance accuracy. The present study explores the performance of BiLSTMs in comparison of Stacked LSTMs, using stock price data from 10 companies listed on the National Stock Exchange of India. It exhibits the effect of bidirectional training in enhancing model precision. The results demonstrate that BiLSTMs, with their advanced training capabilities, provide significantly more accurate stock price forecasts compared to basic structure of LSTMs. However, it was also observed that BiLSTMs take longer to achieve stability compared to their unidirectional LSTM counterparts.

LoRaWAN technology’s reliability is challenged by weather parameters, which can influence the communication channel design, especially when dealing with outdoor devices. We propose to analyze this effect by evaluating the relationship between the received signal strength indicator (RSSI) and different weather parameters, as well as its temporal changes. A rigorous statistical analysis of the RSSI sequences is conducted to assess if they could be represented by a specific statistical model. For this purpose, several models are investigated. The Artificial Intelligence (AI) algorithms cover machine learning (ML) and deep learning methods are appealing when dealing with time series forecasting. Nevertheless, the classical autoregressive integrated moving average (ARIMA) model can be an attractive alternative due to its simplicity. Therefore, this work proposes a comparative study of ARIMA, AI, and hybrid approaches to forecast the RSSI using weather parameters as regressors. The considered AI algorithms are the artificial neural network (ANN), support vector machine (SVM), random forest (RF), and Long Short-Term Memory (LSTM). Also, hybrid models are constructed, coupling the ARIMA with them. The models are evaluated in time series of RSSI, measured by eight different LoRaWAN transmitter nodes and considering the temperature, pressure, relative humidity, and rain as weather parameters. Our analysis reveals that temperature is the dominant factor among weather parameters, and negatively affects RSSI. The ARIMA model that uses only the temperature as a regressor provides consistently better fits than the ARIMA without regressors. Moreover, coupling the ARIMA with the temperature as a regressor and the ANN (ARIMA-ANN) is the best option among the pure AI and hybrid approaches. However, it provided accuracy measures very close to those obtained from the ARIMA model fitted in the first stage, with similar performance. Therefore, the ARIMA model considering the temperature is the most competitive alternative when analyzing RSSI measurements, with the advantage of being the most straightforward method. These results suggest that the RSSI from the analyzed LoRaWAN receiver nodes may not present nonlinear patterns and, considering several weather parameters, they are affected mainly by the outdoor temperature.

Time series forecasting has been used in research and applications in a number of domains such as environmental forecasting, healthcare, finance, supply chain management, and energy consumption. Accurate prediction of future values is necessary for strategic planning operational efficiency and well-informed decision-making regarding time-dependent variables. A hybrid time series forecasting architecture is proposed that combines the strengths of machine learning and statistical models, in particular Gradient Boosting Machines (GBM), Auto-Regressive Integrated Moving Average (ARIMA), and Long Short-Term Memory (LSTM) networks. While LSTM networks and GBM are able to capture complex dependencies and nonlinear patterns, the ARIMA model captures the linear components within the time series. The hybrid model exploits ARIMA interpretability, LSTM temporal memory ability, and GBM ensemble learning efficiency by integrating these three models. Comprehensive experiments conducted on benchmark data sets have shown that the accuracy and reliability of predictions of the proposed hybridization significantly exceeds both individual models and traditional baseline models. The results show that for a variety of real-world applications, hybrid architectures can deliver reliable and accurate time series predictions.

The paper analyzes the long-term noise monitoring data using the AutoRegressive Integrated Moving Average (ARIMA) modeling technique. Box-Jenkins ARIMA approach has been adapted to simulate the daily mean LDay (06-22 h) and LNight (22-06 h) in A- and C-weightings in conjunction with single-noise metrics, day-night average sound level (DNL) for a period of 6 months. The autocorrelation function (ACF) and partial autocorrelation function (PACF) have been obtained to find suitable orders of autoregressive (p) and moving average (q) parameters for ARIMA (p, d, q) models so developed for all the single-noise metrics. The ARIMA models, namely, ARIMA(0,0,14), ARIMA(0,1,1), ARIMA(7,0,0), ARIMA(1,0,0) and ARIMA(0,1,14), have been developed as the most suitable for simulating and forecasting the daily mean LDay dBA, LNight dBA, LDay dBC, LNight dBC, and day-night average sound level (DNL) respectively. The performance of the model so developed is ascertained using the statistical tests. The work reveals that the ARIMA approach is reliable for time-series modeling of traffic noise levels.

Network traffic classification plays a critical role in cybersecurity, quality of service (QoS) management, and anomaly detection. Traditional rule-based classification methods struggle with the increasing complexity and volume of network traffic, necessitating the adoption of machine learning (ML) techniques. In this study, we explore the effectiveness of ML models in classifying network traffic using the NetML dataset, a benchmark dataset that captures diverse traffic patterns, including benign and malicious activities. We preprocess the dataset by applying feature selection, normalization, and data balancing techniques to optimize model performance. Several ML models, including traditional classifiers such as Random Forest (RF), Support Vector Machines (SVM), and K-Nearest Neighbors (KNN), as well as deep learning models such as Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, are trained and evaluated. Model performance is assessed using accuracy, precision, recall, F1- score, and AUC-ROC metrics. Experimental results demonstrate that deep learning models, particularly LSTM networks, achieve superior performance in capturing temporal dependencies in network traffic, significantly outperforming traditional classifiers. Our results indicate that LSTM, GRU, and CNN models all achieved an accuracy of 92.26%, highlighting their effectiveness in network traffic classification. Additionally, feature selection techniques improved computational efficiency without compromising classification performance. However, confusion matrix analysis revealed that the models tend to predict the most frequent class, leading to potential bias and lower accuracy for minority classes. The study also highlights the presence of high values in the confusion matrices, exceeding 70,000 in some cases, indicating dataset imbalance and model bias toward dominant classes. Despite achieving high accuracy, misclassification challenges persist, particularly in identifying encrypted traffic and polymorphic attacks. Transformer-based models demonstrated resilience to adversarial modifications but required significantly higher computational resources. Future work should explore adversarial training, self-supervised learning, and hybrid CNN-LSTM architectures to enhance robustness against evolving cyber threats. Additionally, feature selection optimization and hyperparameter tuning can further refine classification performance, ensuring more reliable deployment in real-world cybersecurity applications.

A comparative analysis of K-Nearest Neighbor, Genetic, Support Vector Machine, Decision Tree, and Long Short Term Memory algorithms in machine learning

Time series analysis plays an important role in many fields. Autoregressive Integrated Moving Average (ARIMA) and Long Short-Term Memory (LSTM) methods are the most common tools to forecast sequential and time series data. However, both of them have obvious defects. The ARIMA model cannot capture non-linear information in sequences. Although the LSTM network is good at learning dynamic variations, it is prone to overfitting and requires a mass of long-term data. In this study, a hybrid technology blending the ARIMA and the LSTM algorithms was utilized to forecast the number of cases of influenza in China. This hybrid method leveraged the advantages of the ARIMA model and the LSTM network. Firstly, the ARIMA model was used to analyze the linear relationship within the time series. Then, the residuals of the ARIMA were taken as the input values to train the LSTM networks. A new hybrid ARIMA-LSTM model obtained, combining a SARIMA (0,1,0) (1,0,0)52 model and a LSTM model with 50 epochs and 32 batch size. This model successfully addressed the previously mentioned issues and enhanced the precision of the predictions. It managed to reduce 4.6% of RMSE, 8.9% of MSE, and 13.9% of MAE. In addition, this new algorithm was found that it didn’t have a high requirement like the individual LSTM model. Since there were not very many observations in the dataset, the performance of the individual LSTM model was not good. However, the integrated model improved this problem and obtained a more precise prediction. Even though the hybrid model had a better performance on prediction, it still has the risk of overfitting the data. The future work will be to improve the hybrid model to decrease this risk by adding more variables and modifying the structure of the LSTM model. Meanwhile, applying this method to another field further proves its feasibility and provides more effective prediction models.

Well production forecasting based on ARIMA-LSTM model considering manual operations

A short-term load forecasting model of multi-scale CNN-LSTM hybrid neural network considering the real-time electricity price

This article is dedicated to the development and research of an advanced hybrid machine learning method for time series forecasting in decision support systems (DSS). The relevance of the work is driven by the rapid growth of data volumes in modern information systems, particularly in cloud infrastructures, and the need for accurate forecasting tools for effective resource management. The objective of the study is to increase the accuracy of computing resource load forecasting by developing a hybrid model that combines the advantages of statistical methods and deep learning architectures. A novel hybrid architecture is proposed, integrating the Autoregressive Integrated Moving Average (ARIMA) model for modeling linear components of a time series, and a Long Short-Term Memory (LSTM) recurrent neural network with a built-in Attention Mechanism for analyzing non-linear residuals. The ARIMA model is used to capture stationary dependencies and seasonality, while the LSTM network with an attention mechanism effectively models complex, non-linear, and long-term patterns in the data remaining after ARIMA processing. An experimental study was conducted on a real dataset of CPU utilization monitoring from virtual machines in the AWS (Amazon Web Services) cloud environment. The proposed hybrid ARIMA-LSTM model with an attention mechanism demonstrated a significant improvement in forecasting accuracy compared to baseline models: pure ARIMA, pure LSTM, and a standard hybrid ARIMA-LSTM model. The Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) metrics for the developed model were 12-18% lower than those of the best-performing baseline model. Scientific novelty lies in the enhancement of existing hybrid approaches by integrating an attention mechanism into the LSTM architecture for analyzing time series residuals. Practical significance of the work consists in the potential for implementing the developed method in automated DSS for optimizing resource allocation, auto-scaling, preventing overloads, and reducing operational costs of cloud infrastructure.