LSTM Research Articles

Machine Learning-Based Financial Big Data Analysis and Forecasting: From Preprocessing to Deep Learning Models

Abstract: Machine learning technology has revolutionized the financial sector by allowing faster and more accurate analysis and forecasting of large-scale financial data. This paper focuses on how machine learning ( ML ), especially deep learning models, can help to deal with high dimensional, noisy, and non-stationary financial data. Essential methodologies such as data preprocessing, feature engineering, and dimensionality reduction are imperative for preparing the raw financial data to ML algorithms. Methods such as outlier detection, normalisation for preprocessing, and feature (variable) selection for dimensionality reduction improve the models accuracy and efficiency. The paper also examines how deep learning models, such as Recurrent Neural Networks ( RNN ) and Long Short-Term Memory ( LSTM ) networks, can overcome the issues of autoregressive integrated moving average ( ARIMA ) models for financial time series prediction. An in-depth comparison of the machine learning models, ranging from supervised to unsupervised methods, is also provided to discuss their pros and cons in the financial domain, including popular applications such as credit scoring, fraud detection, and market risk prediction. The study finally concludes by discussing how optimisation methods such as hyperparameter tuning and cross-validation are imperative for ML models in complex financial scenarios to ensure their generalisation capability and avoid overfitting.

Empowering big data analytics through machine learning: Applications, challenges, and future directions

Abstract. This paper explores the transformative role of machine learning (ML) methodologies in big data analytics, highlighting supervised learning, unsupervised learning, and deep neural networks' contributions to diverse sectors. Supervised learning, with regression analysis at its core, provides accurate forecasting in finance and healthcare by modeling relationships between variables. Unsupervised learning, through techniques like k-means and hierarchical clustering, uncovers patterns within data, offering insights for retail and biological analysis. Deep learning, particularly Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM), excel in complex tasks like image recognition and sequential data processing, driving advances in fields from autonomous driving to language translation. The application of ML in enhancing business intelligence, innovating fintech, and advancing healthcare analytics is discussed, alongside the challenges of data privacy, security, ethical considerations, and the skills gap. This paper underscores the need for advanced cryptographic techniques, bias mitigation strategies, and education to address these challenges. It concludes by emphasizing the importance of interdisciplinary education and AI advancements in bridging the skills gap, ensuring the ethical use of ML, and making these technologies accessible for future innovations.

Big Data Analytics for Forecasting Tourism Recovery in Bali Island Using Multivariate Time Series

Bali is a famous tourist area and can significantly contribute to the Indonesian tourism sector. The COVID-19 pandemic has made Indonesian tourism, including Bali tourism, experience a decline. In March 2022, COVID-19 cases decreased, and the government began to relax some policies. The tourism sector is vital in economic recovery efforts after the COVID-19 pandemic. Therefore, it is necessary to identify tourism recovery to determine strategies and policies related to Indonesian tourism, especially in Bali. Multivariate time series forecasting of tourism demand can be used to identify tourism recovery using several significant data sources. The methods used are Vector Autoregressive (VAR), Support Vector Regression (SVR), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). The data used are the monthly official number of tourists, room occupancy rate, Google Trends, number of booking.com user reviews, and nighttime light intensity in Bali Province from January 2019 to December 2022. The results show that the best forecasting method is VAR, and modeling with multivariate time series forecasting can improve the performance of forecasting results. In addition, big data can be used as a source of supporting data that can provide better forecasting results, and the size of the dataset affects the selection of the best model. Furthermore, the descriptive and forecasting analysis results show that Bali tourism has experienced post-pandemic tourism recovery. The strategies and policies of the Bali government to restore Bali tourism faster are good enough.

Classification of hand movements from EEG using a FusionNet based LSTM network.


Accurate classification of electroencephalogram (EEG) signals is crucial for advancing brain-computer interface (BCI) technology. However, current methods face significant challenges in classifying hand movement EEG signals, including effective spatial feature extraction, capturing temporal dependencies, and representing underlying signal dynamics.
Approach.
This paper introduces a novel multi-model fusion approach, FusionNet-LSTM, designed to address these issues. Specifically, it integrates Convolutional Neural Networks (CNN) for spatial feature extraction, Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM) networks for capturing temporal dependencies, and Autoregressive (AR) models for representing signal dynamics.
Main results.
Compared to single models and state-of-the-art methods, this fusion approach demonstrates substantial improvements in classification accuracy. Experimental results show that the proposed model achieves an accuracy of 87.1% in cross-subject data classification and 99.1% in within-subject data classification. Additionally, Gradient Boosting Trees were employed to evaluate the significance of various EEG features to the model.
Significance.
This study highlights the advantages of integrating multiple models and introduces a superior classification model, which is pivotal for the advancement of BCI systems.

LSTM Algorithm in Predicting Chronic Kidney Disease Optimized Using Genetic Algorithm

Chronic Kidney Disease is a health condition in which the kidneys experience a progressive decline in function. Kidneys are vital organs that filter waste and excess blood fluids. CKD can lead to excess products in the body and cause various health issues, so early detection of CKD is necessary. While traditional machine learning techniques have performed well in predicting CKD in existing studies, this study investigates the potential of long short-term memory (LSTM) optimized with Genetic Algorithm to enhance predictive accuracy and efficiency by optimizing its hyperparameters, including number of units, hidden layers, activation function, recurrent activation, and dropout rate. The result demonstrates that the optimized LSTM slightly performs better than without optimization, achieving higher precision, recall, accuracy, and f1 score by 100% respectively. This outstanding result can be attributed to several key factors, such as ensuring rigorous data preprocessing and utilizing k-fold cross-validation to make the model more reliable. This indicates the hybrid approach can be a powerful method for the early detection of CKD, leading to better patient outcomes. Despite the promising performance, further research is suggested, specifically using a larger dataset to ensure applicability to more general population and exploring other optimization methods to reduce computational cost.

Empirical Modal Decomposition Combined with Deep Learning for Photoacoustic Spectroscopy Detection of Mixture Gas Concentrations.

In photoacoustic spectroscopy based multicomponent gas analysis, the overlap of the absorption spectra among different gases can affect the measurement accuracy of gas concentrations. We report a multicomponent gas analysis method based on empirical modal decomposition (EMD), convolutional neural networks (CNN), and long short-term memory (LSTM) networks that can extract the exact concentrations of mixed gases from the overlapping wavelength-modulated spectroscopy with second harmonic (WMS-2f) detection. The WMS-2f signals of 25 different concentration combinations of acetylene-ammonia mixtures are detected using a single distributed feedback laser (DFB) at 1531.5 nm. The acetylene concentrations range from 2.5 to 7.5 ppm and the ammonia concentrations from 12.5 to 37.5 ppm. The data set is enhanced by cyclic shifting and adding Gaussian noise. The classification accuracy of the test set reaches 99.89% after tuning. The mean absolute errors of the five additional sets of data measured under different conditions are 0.092 ppm for acetylene and 1.902 ppm for ammonia, within the above concentration ranges.

Time Series Analysis and Optimization of the Prediction Model of Agricultural Insurance Loss Ratio

For ensuring successful financial planning to perform sustainable farming, one key sector is to provide solutions that could accurately predict the agricultural loss ratios. In China, the Henan province is considered to be an agricultural center that is primarily exposed to drastic weather fluctuations that directly impact the crop yields. This study was conducted in Henan province from January 2020 to December 2023. With the data collected from that period, the study proposes a combinatory model combining Deep Gaussian Processes with Bayesian Long Short-Term Memory (LSTM) networks. The model was trained on data encompassing weather conditions, agricultural practices, and historical insurance claims. The experimental analysis was conducted against other traditional models, including ARIMA and Support Vector Regression. The RMSE improvement of the proposed model was around 7.2% on training data and 8.2% on test data, which demonstrates enhanced predictive accuracy. The enhanced performance of the proposed model was reflected in its effectiveness in reducing log-likelihood errors across training epochs. The model had demonstrated better robustness in handling complex and multi-dimensional agricultural data.

Multi-Timescale Energy Consumption Management in Smart Buildings Using Hybrid Deep Artificial Neural Networks

Demand side management is a critical issue in the energy sector. Recent events such as the global energy crisis, costs, the necessity to reduce greenhouse emissions, and extreme weather conditions have increased the need for energy efficiency. Thus, accurately predicting energy consumption is one of the key steps in addressing inefficiency in energy consumption and its optimization. In this regard, accurate predictions on a daily, hourly, and minute-by-minute basis would not only minimize wastage but would also help to save costs. In this article, we propose intelligent models using ensembles of convolutional neural network (CNN), long-short-term memory (LSTM), bi-directional LSTM and gated recurrent units (GRUs) neural network models for daily, hourly, and minute-by-minute predictions of energy consumptions in smart buildings. The proposed models outperform state-of-the-art deep neural network models for predicting minute-by-minute energy consumption, with a mean square error of 0.109. The evaluated hybrid models also capture more latent trends in the data than traditional single models. The results highlight the potential of using hybrid deep learning models for improved energy efficiency management in smart buildings.

A Hybrid Quantum-Classical Model for Stock Price Prediction Using Quantum-Enhanced Long Short-Term Memory

The stock markets have become a popular topic within machine learning (ML) communities, with one particular application being stock price prediction. However, accurately predicting the stock market is a challenging task due to the various factors within financial markets. With the introduction of ML, prediction techniques have become more efficient but computationally demanding for classical computers. Given the rise of quantum computing (QC), which holds great promise for being exponentially faster than current classical computers, it is natural to explore ML within the QC domain. In this study, we leverage a hybrid quantum-classical ML approach to predict a company’s stock price. We integrate classical long short-term memory (LSTM) with QC, resulting in a new variant called QLSTM. We initially validate the proposed QLSTM model by leveraging an IBM quantum simulator running on a classical computer, after which we conduct predictions using an IBM real quantum computer. Thereafter, we evaluate the performance of our model using the root mean square error (RMSE) and prediction accuracy. Additionally, we perform a comparative analysis, evaluating the prediction performance of the QLSTM model against several other classical models. Further, we explore the impacts of hyperparameters on the QLSTM model to determine the best configuration. Our experimental results demonstrate that while the classical LSTM model achieved an RMSE of 0.0693 and a prediction accuracy of 0.8815, the QLSTM model exhibited superior performance, achieving values of 0.0602 and 0.9736, respectively. Furthermore, the QLSTM outperformed other classical models in both metrics.

TRANSFORMING BANKING SECURITY: THE ROLE OF DEEP LEARNING IN FRAUD DETECTION SYSTEMS

In the digital banking landscape, the increasing volume of online transactions has heightened the risk of fraudulent activities, necessitating the development of more effective detection systems. This study investigates the efficacy of various machine learning and deep learning algorithms in identifying fraudulent transactions, emphasizing Long Short-Term Memory (LSTM) networks. We implemented and evaluated multiple algorithms, including Logistic Regression, Random Forest, Gradient Boosting Machines (GBM), and XGBoost, on a large-scale credit card transaction dataset. Our results demonstrate that the LSTM model outperforms traditional machine learning algorithms, achieving an accuracy of 98.5%, precision of 87.2%, recall of 85.0%, and an Area Under the Curve (AUC) score of 0.94. These findings highlight the superior capability of LSTM networks to capture complex patterns in sequential transaction data, making them an asset for real-time fraud detection in banking. This research underscores the need for financial institutions to adopt advanced deep learning techniques to enhance their fraud detection systems, thereby minimizing financial losses and improving customer trust.

Biomechanical and machine learning approaches to automating the identification of musical styles and emotions through human motion analysis

This study explores the intricate relationship between biomechanical movements and musical expression, focusing on the identification of musical styles and emotions. Violin performance is characterized by complex interactions between physical actions—such as bowing techniques, finger placements, and posture—and the resulting acoustic output. Recent advances in motion capture technology and sound analysis have enabled a more objective examination of these processes. However, the current literature frequently addresses biomechanics and acoustic features in isolation, lacking an integrated understanding of how physical movements translate into specific musical expressions. Machine Learning (ML), particularly Long Short-Term Memory (LSTM) networks, provides a promising avenue for bridging this gap. LSTM models are adept at capturing temporal dependencies in sequential data, making them suitable for analyzing the dynamic nature of violin performance. In this work, they have proposed a comprehensive model that combines biomechanical analysis with Mel-spectrogram-based LSTM modeling to automate the identification of musical styles and emotions in violin performances. Using motion capture systems, Inertial Measurement Units (IMUs), and high-fidelity audio recordings, we collected synchronized biomechanical and acoustic data from violinists performing various musical excerpts. The LSTM model was trained on this dataset to learn the intricate connections between physical movements and the acoustic features of each performance. Key findings from the study demonstrate the effectiveness of this integrated approach. The LSTM model achieved a validation accuracy of 92.5% in classifying musical styles and emotions, with precision, recall, and F1-score reaching 94.3%, 92.6%, and 93.4%, respectively, by the 100th epoch. The analysis also revealed strong correlations between specific biomechanical parameters, such as shoulder joint angle and bowing velocity, and acoustic features, like sound intensity and vibrato amplitude.

Research on Climactic Chapter Recognition of a Chinese Long Novel Based on Plot Description

Many readers continue to pursue Chinese long novels in the past several decades because of diverse characters and fascinating plots. The climactic chapter is an important part of a Chinese long novel, where the key conflict develops to the extreme point. However, how to quickly and accurately recognize the climactic chapter remains a common problem for many readers in their reading choices. This paper conducts research on recognizing the climactic chapter of a Chinese long novel by accurately describing its plot. The proposed method consists of two parts; one is the extraction of key elements, such as viewpoint paragraphs, non-viewpoint paragraphs, chapter keywords, major characters etc. The other part is the climactic chapter recognition, which applies the Bidirectional Gate Recurrent Unit (BiGRU) model and the multi-head attention to recognize the climactic chapter, on the basis of the chapter plot description matrix. Comparative experiments on the corpus named The Condor Trilogy show that the proposed method in this paper has a better recognition performance compared with the existing models, such as Naive Bayesian (NB), Support Vector Machine (SVM), Roberta-large, and the Bidirectional Long-Short Term Memory (BiLSTM) network. Ablation experiments validated the effectiveness of primary components in the proposed method.

Advancing smart city factories: enhancing industrial mechanical operations via deep learning techniques

In the contemporary realm of industry, the imperative for influential and steadfast systems to detect anomalies is critically recognized. Our study introduces a cutting-edge approach utilizing a deep learning model of the Long-Short Term Memory variety, meticulously crafted for real-time surveillance and mitigation of irregularities within industrial settings. Through the careful amalgamation of data acquisition and analytic processing informed by our model, we have forged a system adept at pinpointing anomalies with high precision, capable of autonomously proposing or implementing remedial measures. The findings demonstrate a marked enhancement in the efficacy of operations, with the model’s accuracy surging to 95%, recall at 90%, and an F1 score reaching 92.5%. Moreover, the system has favorably impacted the environment, evidenced by a 25% decline in CO2 emissions and a 20% reduction in water usage. Our model surpasses preceding systems, showcasing significant gains in speed and precision. This research corroborates the capabilities of deep learning within the industrial sector. It underscores the role of automated systems in fostering more sustainable and efficient operations in the contemporary industrial landscape.

Optimised Deep Learning for Time-Critical Load Forecasting Using LSTM and Modified Particle Swarm Optimisation

Short-term electric load forecasting is critical for power system planning and operations due to demand fluctuations driven by variable energy resources. While deep learning-based forecasting models have shown strong performance, time-sensitive applications require improvements in both accuracy and convergence speed. To address this, we propose a hybrid model that combines long short-term memory (LSTM) with a modified particle swarm optimisation (mPSO) algorithm. Although LSTM is effective for nonlinear time-series predictions, its computational complexity increases with parameter variations. To overcome this, mPSO is used for parameter tuning, ensuring accurate forecasting while avoiding local optima. Additionally, XGBoost and decision tree filtering algorithms are incorporated to reduce dimensionality and prevent overfitting. Unlike existing models that focus mainly on accuracy, our framework optimises accuracy, stability, and convergence rate simultaneously. The model was tested on real hourly load data from New South Wales and Victoria, significantly outperforming benchmark models such as ENN, LSTM, GA-LSTM, and PSO-LSTM. For NSW, the proposed model reduced MSE by 91.91%, RMSE by 94.89%, and MAPE by 74.29%. In VIC, MSE decreased by 91.33%, RMSE by 95.73%, and MAPE by 72.06%, showcasing superior performance across all metrics.

A groundwater level spatiotemporal prediction model based on graph convolutional networks with a long short-term memory

ABSTRACT The performance of regional groundwater level (GWL) prediction model hinges on understanding intricate spatiotemporal correlations among monitoring wells. In this study, a graph convolutional network (GCN) with a long short-term memory (LSTM) (GCN–LSTM) model is introduced for GWL prediction utilizing data from 16 wells located in the northeastern Xiangtan City, China. This model is designed to account for both the hybrid temporal dependencies and spatial autocorrelations among wells. It consists of two parts: the spatial part employs GCNs to extract spatial characteristics from a spatial self-similarity weight matrix and an attribute self-similarity wight matrix among wells; the temporal part utilizes a LSTM module to capture the temporal patterns of GWL sequences, along with monthly precipitation and temperature data. This model dynamically predicts changes in groundwater levels, achieving higher accuracy on average compared to single-well predictions using LSTM. By incorporating both temporal dependencies and spatial autocorrelations, the GCN–LSTM model demonstrated an average improvement in goodness-of-fit of approximately 11.21% over the LSTM-based model for individual wells. Its application holds significant reference value for the sustainable utilization and development of groundwater resources in Xiangtan City.

Long‐Term Estimation of SoH Using Cascaded LSTM‐RNN for Lithium Batteries Subjected to Aging and Accelerated Degradation

ABSTRACTAccurate estimation of state of health (SoH) of the battery over long‐term is a critical challenge for the battery management systems in electric vehicles. This is due to the challenges in accurately modeling the accelerated aging and degradation phenomena caused by diverse operating conditions of the battery. This paper presents a cascaded recurrent neural networks (RNN) with long short‐term memory (LSTM) to estimate the internal resistance and SoH, taking account of various abnormal operating conditions of the battery. A datasheet‐based degradation model of the battery is developed using fade equations. The training and validation data set for LSTM‐RNN are generated by subjecting the battery model to various factors that cause accelerated degradation, such as fast charging, varying operating temperatures, overutilization, and cell imbalance. The cascaded LSTM‐RNN is trained to estimate SoH only once after the completion of every charge–discharge cycle. The training error index parameters of the proposed SoH estimator are well within 1%, demonstrating the reliability and robustness of the estimator to diverse operating conditions of the battery.

Machine learning model to study effects of lightweight racket design on elbow and shoulder loading during tennis serve

Tennis is a famous sport in which players perform high-speed, repetitive movements that cause significant load to the shoulder and elbow joints, particularly during tennis serve. Further, the racket design also plays a key role in the efficiency of the player’s performance through the level of mechanical stress that it places on the player’s elbow joints. Therefore, analyzing the biomechanical impact of racket types on the load it renders on elbow and shoulder loading will help optimize the player’s performance and avoid the possibility of risk. However, there are limited studies that are related to the effect of differences in racket mass, balance, and inertia over joint force and moment during the phases of serve. To address this gap, this study employs a Machine Learning (ML)-based model to impact three types of racket such as head-light, even-balanced, and head-heavy, towards joint moments and forces in the shoulder, elbow, and wrist during tennis serves. The kinematic data was collected from eight tennis players, and the collected data was processed using a Long Short-Term Memory (LSTM) neural network to predict joint moments and forces based on racket parameters and segmental kinematics. The results have shown that racket design has an excellent impact on a player’s performance through its impact on shoulder and elbow joints. The head-light racket resulted in a shoulder adduction moment of 9.4 ± 0.8 Nm and a shoulder joint force of 130.2 ± 9.4 N during the acceleration phase, compared to the head-heavy racket, which generated a higher adduction moment of 11.3 ± 1.0 Nm and a shoulder force of 162.3 ± 11.4 N. The even-balanced racket showed intermediate values, with an adduction moment of 10.2 ± 0.9 Nm and a shoulder force of 142.5 ± 9.9 N.

An improved prediction method for BDS-3 SISA parameters and the preliminary performance evaluation

Abstract The ephemeris accuracy parameters broadcast in the ephemeris messages are commonly utilized as an alarm threshold for the signal-in-space ranging errors (SISRE) of the Global Navigation Satellite System (GNSS). Accurately predicting the orbit and clock errors of broadcast ephemeris, and converting the prediction errors into broadcast ephemeris accuracy parameters using a robust model, is important for the integrity service of GNSS. The BeiDou Navigation Satellite System (BDS-3) broadcasts four ephemeris accuracy parameters, namely SIS A oe , SIS A ocb , SIS A oc1 , and SIS A oc2 , to represent the standard deviations of the satellite orbit and clock prediction errors in the broadcast ephemeris. The probability that the broadcast ephemeris SISRE exceeds 4.42 times the signal-in-space accuracy (SISA) without triggering an alarm in time is expected to be less than 1 × 10−5. In addition to accurately enveloping the broadcast ephemeris SISRE, SISA should also reduce the over-envelopment margin over SISRE to accurately express the system service accuracy. Thus, an accurate prediction of orbit and clock errors in the broadcast ephemeris is crucial for calculating the SISA. An orbit and clock error prediction method based on the Long Short-Term Memory (LSTM) neural network is proposed in this study. This method uses historical satellite orbit and clock errors, derived from the difference between precise and broadcast ephemeris, to train an LSTM neural network. The trained LSTM model is then used to predict the orbit and clock errors for the next 24 h. The predicted errors are used to calculate the corresponding SISA for the broadcast ephemeris. The results indicate that the 24 h accuracies of the predicted orbit and clock errors are less than 0.06 m and 0.25 m, respectively. The calculated SISA can completely envelop the actual broadcast ephemeris SISRE, achieving 100% coverage while also decreasing the over-envelopment margin by 80%. Furthermore, the horizontal protection level and vertical protection level calculated using the corresponding SISA are reduced by approximately 14%, meeting the integrity requirements set by the International Civil Aviation Organization for Approach with Vertical Guidance I. The proposed method enhances the timeliness of SISA parameter calculation and the integrity and availability of the BDS-3 service, serving as a crucial reference value for safety critical applications such as civil aviation services.

Using DL Models in the Service Layer to Enhance the Fault Tolerance of IoT Networks

In an IoT network, the networked servers form a service layer, providing services to the users and the devices. The request to the service servers is routed through the gateway on one side of the services layer and the networked controllers on the other side. Data are transported from the sensors/devices through cluster heads en route to base stations and the controllers to the service servers, where the data are processed and sent for storage in the cloud through gateways. When any device is broken down or becomes non-operational, the inputs are not sensed, creating a gap in the data. The data transmitted from the devices would then become an incomplete flow; such data are not suitable for undertaking data analytics or predictions. The missing data must be first identified as the data flow and estimated or predicated to complete the data before they are transmitted through the cloud for storage and subsequent retrievals. This paper proposes a recurrent (RNN) neural network to predict the missing data. Two models are tested to predict the missing data: the multi-layer perceptron (MLP) model and a long short-term memory (LSTM)-based RNN model. The RNN-based model provides 99.66% accurate data prediction compared to other models.

Prediction and assessment of meteorological drought in southwest China using long short-term memory model

Abstract Drought prediction is crucial for mitigating risks and designing measures to alleviate its impact. Machine learning models have been widely applied in the field of drought prediction in recent years. This study concentrated on predicting meteorological droughts in southwest China, a region prone to frequent and severe droughts, particularly in areas with sparse meteorological station coverage. The long short-term memory (LSTM) predictive model, which is a deep learning model, was constructed by calculating standardized precipitation evapotranspiration index (SPEI) values based on 144 weather station observations from 1980 to 2020. The 5-fold cross-validation method was used for the hyperparameter optimization of the model. The LSTM model underwent comprehensive assessment and validation through multiple methods. This included the use of several accuracy assessment indicators and a comparison of results. The comparison covered different drought characteristics among the LSTM predictive model, the benchmark random forest (RF) predictive model, the historical drought situations, and the calculated SPEI values based on observations from 144 weather stations. The results showed that the training results of the LSTM predictive model basically agreed with the SPEI values calculated from weather station observations. The model-predicted variation trend of SPEI values for 2020 was similar to the variation in SPEI values calculated based on weather station observations. On the test set, the coefficient of determination (R 2), the root mean square error, the explained variance score, the Nash–Sutcliffe efficiency, and the Kling–Gupta efficiency were 0.757, 0.210, 0.802, 0.761, and 0.212, respectively. The total consistency rate of the drought grade was 59.26%. The spatial correlation distribution of SPEI values between LSTM model prediction and calculation from meteorological stations in 2020 was more than 0.5 for most regions. The correlation coefficients exceeded 0.6 in western Tibet and Chengdu Plains. Compared to the RF model, the LSTM model excelled in all five performance evaluation metrics and demonstrated a higher overall consistency rate for drought categories. The Kruskal–Wallis test for both the LSTM and RF models all indicated no significant difference in the distributions between the predicted and observed data. Scatter plots revealed that the prediction accuracy for both models in 2020 was suboptimal, with the SPEI showing a comparatively narrow range of values. Nonetheless, the LSTM model significantly outperformed the RF model in terms of prediction accuracy. In summary, the LSTM model demonstrated good overall performance, accuracy, and applicability. It has the potential to enhance dynamic drought prediction in regions with complex terrain, diverse climatic factors, and sparse weather station networks.