LSTM Research Articles

A testbed for landslide prediction with blockchain-based transmission and cloud offloading

Purpose The purpose of this paper is to develop an IoT-based testbed for land displacement monitoring in real-time with blockchain-enabled transmission and machine learning for predictions. Cloud offloading has also been incorporated into the system proposed. Design/methodology/approach The system consists of a modelled landslide testbed at a laboratory scale with soil, water level, humidity, temperature sensors and a designed extensometer using a 3D printer. An Arduino microcontroller handles all sensor information, and a Raspberry Pi performs blockchain and transmission to a gateway using a 4G transmission module. The transmitted data is received in a server GUI application and a webpage where long short-term memory (LSTM) and multi-layer perceptron (MLP) are used for predictions. For further scalability of this system, cloud offloading was implemented, allowing the information to be accessed across multiple platforms. Findings The accuracy of the designed extensometer was compared to an industrial-grade extensometer. Moreover, the predictions performed with MLP and LSTM yielded a MAPE of 6.0% and 7.8%, respectively. Finally, the blockchain analysis demonstrated that using smaller block sizes provides better security but lower throughput than large block sizes. Originality/value A sophisticated testbed for landslide monitoring, which includes blockchain, AI and cloud offloading, has been proposed along with in-depth performance analysis.

Research of Automatic Modulation Recognition Technology based on Deep Learning

Automatic modulation recognition is a vital component in communication systems. It helps in recognizing and classifying signal modulation automatically without the acquirement of the signals modulation type in advance. This paper aims to provide a comprehensive exploration of the application of deep learning techniques in the field of Automatic Modulation Recognition (AMR), with a specific emphasis on the analysis and comparison of two prominent neural network architectures: Convolutional Neural Networks (CNN) and Convolutional Long Short-Term Memory Networks (CLDNN). By delving into the structural and functional aspects of these models, the paper seeks to elucidate how CNNs, with their ability to capture spatial features through convolutional layers, and CLDNNs, which enhance this capability by integrating Long Short-Term Memory (LSTM) units to handle temporal dynamics, contribute to the advancement of AMR technology. We used the DeepSig dataset: RadioML 2018.01A to evaluate them, comparing their performance in different signal-to-noise ratio scenarios. The results manifested that CLDNN performed better than CNN obviously, especially in the low SNR scenario where CLDNN reached a higher validation set accuracy.

Cryptocurrency Portfolio Optimisation Based on LSTM Time Series Forecasting

Recently, with the gradual development of machine learning technology, more and more people are trying to apply machine learning technology in various fields, and finance is one of the important fields. This work investigates the optimization of cryptocurrency portfolios by combining Long Short-Term Memory (LSTM) time series forecasting with traditional portfolio optimization methods. The focus of the paper is on using the historical price data from the past six years of Bitcoin (BTC), Ethereum (ETH), and Litecoin (LTC) to train LSTM models, which are then used to predict the prices of these cryptocurrencies for the period from January to June 2024. These predictions are subsequently incorporated into an extended Markowitz framework to optimize the portfolio on a monthly basis. The results indicate that the LSTM-enhanced portfolio optimization method yields higher returns and better risk management compared to traditional methods. This finding could prove that it is feasible and effective to apply machine learning methods, especially time series forecasting methods, to cryptocurrency portfolios.

A hybrid model based on CNN-LSTM for assessing the risk of increasing claims in insurance companies

This article proposes a hybrid model to assist insurance companies accurately assess the risk of increasing claims for their premiums. The model integrates long short-term memory (LSTM) networks and convolutional neural networks (CNN) to analyze historical claim data and identify emerging risk trends. We analyzed data obtained from insurance companies and found that the hybrid CNN-LSTM model outperforms standalone models in accurately assessing and categorizing risk levels. The proposed CNN-LSTM model achieved an accuracy of 98.5%, outperforming the standalone CNN (95.8%) and LSTM (92.6%). We implemented 10-fold cross-validation to ensure robustness, confirming consistent performance across different data splits. Furthermore, we validated the model on an external dataset to assess its generalizability. The results demonstrate that the model effectively classifies insurance risks in different market environments, highlighting its potential for real-world applications. Our study contributes to the insurance industry by providing valuable insights for effective risk management strategies and highlights the model’s broader applicability in global insurance markets.

Abstract 2508: A multiscale framework combining mechanistic models and LSTM-GCN networks for predicting drug sensitivity

Abstract Understanding tumor evolution and treatment resistance requires an integrative approach that combines cellular dynamics with molecular interactions. While mechanistic models excel in capturing physiological processes and predicting behaviors beyond the original data, their specificity limits their generalizability. Conversely, deep learning enables universal predictions but often lacks interpretability, posing challenges for application in cancer research. To address these limitations, we propose a novel computational framework that integrates mechanistic simulations, graph-based modeling, Long Short-Term Memory (LSTM) networks, and Bayesian inference to predict the sensitivity of six cancer types to targeted therapies, chemotherapy agents, immunotherapies, and their combinations. In this framework, mechanistic models provide an individualized baseline simulation for different cancer types, while deep learning integrates these outputs with drug-related information to generate robust and interpretable predictions tailored to specific types. Our hybrid framework leverages a mechanistic pan-cancer model that simulates molecular signaling pathways, cell cycle dynamics, and apoptosis across four biological compartments. The model is parameterized using published gene expression profiles and mutation data from xenograft models, including key mutations such as KRAS, EGFR, and TP53. Simulations are performed for six cancer types, including breast carcinoma, colorectal cancer, and pancreatic carcinoma, to approximate patient-specific tumor biology. Tumor cell populations generated from these simulations are represented as graphs, with each node corresponding to an individual cell characterized by proliferation status and dynamic molecular states (e.g., protein concentrations and phosphorylation levels). Edges are weighted and defined based on molecular similarity between cells. The graph is processed using a Graph Convolutional Network (GCN), which generates embeddings that capture tumor population heterogeneity. These embeddings are then fed into an LSTM to extract the temporal dynamics of the tumor population. The final hidden states computed by the LSTM are further combined with embeddings of drug chemical structure and drug-target interactions to make drug sensitivity predictions. To enhance the robustness of predictions, Bayesian inference is incorporated to quantify uncertainty in estimated drug responses. This framework bridges the gap between mechanistic modeling and machine learning, offering a scalable and interpretable approach to studying tumor evolution under the selective power of drug treatment. By integrating the physiological insights of mechanistic models with the predictive power of deep learning, it enables the identification of molecular drivers of resistance and informs the development of personalized therapies. Citation Format: Chenhui Ma, Evren Gurkan-Cavusoglu. A multiscale framework combining mechanistic models and LSTM-GCN networks for predicting drug sensitivity [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2508.

A case study on long short-term memory-based modeling for predicting chlorine and free ammonia levels at water treatment plant testing sites

ABSTRACT This work explores the accurate prediction of water quality, a critical aspect of water environment protection. Leveraging advanced algorithms and statistical models, including regression analyses, moving average, and artificial neural networks (ANN), we analyze multivariate time-series datasets from tap water testing sites. Our approach involves model fitting and evaluation based on the Akaike information criterion (AIC) and root mean square error (RMSE). The work introduces the long short-term memory (LSTM) neural network and the seasonal autoregressive integrated moving average (SARIMA) model, detailing their architectures and implementations. Results highlight the effectiveness of a multivariate LSTM model in forecasting total chlorine concentrations, complemented by the SARIMA model. Evaluation metrics reveal comparable predictive accuracy, with a slight advantage observed for the mixed method combining both models. This work underscores the significance of integrating technological and statistical methodologies for enhanced water quality predictions, contributing to proactive water management strategies.

Forced Oscillation Detection via a Hybrid Network of a Spiking Recurrent Neural Network and LSTM

The detection of forced oscillations, especially distinguishing them from natural oscillations, has emerged as a major concern in power system stability monitoring. Deep learning (DL) holds significant potential for detecting forced oscillations correctly. However, existing artificial neural networks (ANNs) face challenges when employed in edge devices for timely detection due to their inherent complex computations and high power consumption. This paper proposes a novel hybrid network that integrates a spiking recurrent neural network (SRNN) with long short-term memory (LSTM). The SRNN achieves computational and energy efficiency, while the integration with LSTM is conducive to effectively capturing temporal dependencies in time-series oscillation data. The proposed hybrid network is trained using the backpropagation-through-time (BPTT) optimization algorithm, with adjustments made to address the discontinuous gradient in the SRNN. We evaluate our proposed model on both simulated and real-world oscillation datasets. Overall, the experimental results demonstrate that the proposed model can achieve higher accuracy and superior performance in distinguishing forced oscillations from natural oscillations, even in the presence of strong noise, compared to pure LSTM and other SRNN-related models.

A Multi-Scale Fusion Convolutional Network for Time-Series Silicon Prediction in Blast Furnaces

In steel production, the blast furnace is a critical element. In this process, precisely controlling the temperature of the molten iron is indispensable for attaining efficient operations and high-grade products. This temperature is often indirectly reflected by the silicon content in the hot metal. However, due to the dynamic nature and inherent delays of the ironmaking process, real-time prediction of silicon content remains a significant challenge, and traditional methods often suffer from insufficient prediction accuracy. This study presents a novel Multi-Scale Fusion Convolutional Neural Network (MSF-CNN) to accurately predict the silicon content during the blast furnace smelting process, addressing the limitations of existing data-driven approaches. The proposed MSF-CNN model extracts temporal features at two distinct scales. The first scale utilizes a Convolutional Block Attention Module, which captures local temporal dependencies by focusing on the most relevant features across adjacent time steps. The second scale employs a Multi-Head Self-Attention mechanism to model long-term temporal dependencies, overcoming the inherent delay issues in the blast furnace process. By combining these two scales, the model effectively captures both short-term and long-term temporal dependencies, thereby enhancing prediction accuracy and real-time applicability. Validation using real blast furnace data demonstrates that MSF-CNN outperforms recurrent neural network models such as Long Short-Term Memory (LSTM) and the Gated Recurrent Unit (GRU). Compared with LSTM and the GRU, MSF-CNN reduces the Root Mean Square Error (RMSE) by approximately 22% and 21%, respectively, and improves the Hit Rate (HR) by over 3.5% and 4%, highlighting its superiority in capturing complex temporal dependencies. These results indicate that the MSF-CNN adapts better to the blast furnace’s dynamic variations and inherent delays, achieving significant improvements in prediction precision and robustness compared to state-of-the-art recurrent models.

Enhancing water saturation predictions from conventional well logs in a carbonate gas reservoir with a hybrid CNN-LSTM model

Predicting water saturation in the South Pars gas field’s carbonate reservoirs has been challenging due to limitations in conventional methods like the Archie equation. This leads to unproduced gas volume and water coning phenomena, leaving a gap for an intelligent model to determine the water saturation profile more accurately. Four machine learning algorithms: XGBoost, long short-term memory (LSTM), 1-dimensional convolutional neural network (1D-CNN), and a hybrid CNN-LSTM were developed to predict experimental water saturation values from conventional well logs. A total of 10,674 data points were collected from four wells in the South Pars gas field, where well logging evaluations and core measurements were available. The methodology includes data pre-processing to de-noise and transform data into a proper format, feature engineering followed by feature selection based on domain knowledge, models’ training, testing, and optimizing. XGBoost hyperparameters were tuned using grid search, and each neural network model was optimized using a genetic algorithm. The network architectures were designed with specific layers and activation functions to enhance predictive capabilities. A robust validation process was conducted using a blind well dataset. Archie equation was also applied for better comparative analysis. Results showed that the CNN-LSTM model was the most effective, with an R-squared accuracy score of 0.859, while the Archie equation had the least R-squared accuracy of 0.322 in the validation dataset. Furthermore, utilizing an inclusive dataset, a new CNN-LSTM model was constructed with a 10-fold framework, achieving an average R-squared of 0.968. This model was utilized to develop an application that displays the reservoir’s water saturation profile and other features as a well log report within a dashboard. The findings support two points: first, 1D-CNN and LSTM work better together in a hybrid structure, and second, the Archie equation is ineffective in carbonate reservoirs of the South Pars gas field. The study intends to enrich the literature by employing a hybrid machine learning algorithm compared to earlier studies and surpassing traditional protocols by comparing algorithms in a blind well to offer a reliable validation mechanism.Graphical abstract

Research and Application of Gas Drainage Negative Pressure Regulation Method Considering Permeability Differences

This study investigates a dynamic regulation strategy for intelligent gas drainage negative pressure using a comprehensive approach involving numerical simulations, intelligent algorithms, and field experiments. In the numerical simulation component, a permeability evolution model was developed to characterize the area in front of the mining face. Simulations were performed under three negative pressure settings (13 kPa, 18 kPa, and 25 kPa) to investigate the relationships among drainage negative pressure, gas concentration, and flow rate. For the intelligent algorithm, a Long Short-Term Memory (LSTM) prediction model was developed to forecast drainage negative pressure. Based on the predictions, a dynamic regulation strategy for intelligent gas drainage negative pressure was formulated. For field validation, a 120-day in situ experiment was carried out. Intelligent control valves and monitoring instruments were deployed across various sections of the coal seam ahead of the mining face, validating the proposed regulation strategy. The results indicate that permeability is highest in the pressure-relief zone ahead of the mining face and lowest in the stress concentration zone. In the original stress zone, which is unaffected by mining disturbances, the permeability remains unchanged. Drainage negative pressure is positively correlated with gas flow rate, but negatively correlated with gas concentration. In the stress concentration zone, when drainage negative pressure reaches 25 kPa, permeability ceases to be the dominant factor influencing gas flow. At this stage, the pressure gradient between the gas in coal fractures and the drainage system becomes the primary driving force for gas flow. The intelligent dynamic regulation strategy for gas drainage, underpinned by the LSTM prediction model, demonstrated strong performance in field applications. Following intelligent regulation, gas concentrations in various regions showed significant improvement. The findings of this study actively contribute to the advancement of intelligent gas drainage technology.

LSTM-based Portfolio Optimization Study in the Food and Beverage Industry

This study investigates how to best allocate a portfolio using long short-term memory (LSTM) networks to anticipate stock prices in the food and beverage industry. The 10-year historical daily closing prices of five major food and beverage stocks were collected as model training data. A recurrent neural network model with LSTM and fully connected layers was constructed. The weight allocation of different stocks in the initial portfolio was evaluated by Monte Carlo simulation. The Nestle stock with higher risk-return was selected for individual stock price prediction. The results show that Nestle's stock price has a slight upward trend, and it is recommended to increase its weighting in the portfolio appropriately. However, the real range of price variations cannot be correctly anticipated due to the incomplete consideration of influencing factors. The main conclusion is that the LSTM model can reasonably fit the historical stock price time series and judge the general trend, but it needs to expand the data and features to improve the prediction accuracy. This preliminary study provides a basic analysis for LSTM predictive modelling of food and beverage stocks, but further optimization of the model design and expansion of the data range are needed to achieve a reliable optimization of the risk-return trade-off in portfolio management.

Dual-channel knowledge attention for traditional Chinese medicine syndrome differentiation

With the rapid advancement of Natural Language Processing (NLP) technologies, the application of NLP to enable intelligent syndrome differentiation in Traditional Chinese Medicine (TCM) has become a popular research focus. However, TCM texts contain numerous obscure characters and specialized terminologies, which existing methods struggle to effectively extract, leading to lower accuracy in syndrome differentiation. To address this, we propose a dual-channel knowledge-attention model for TCM syndrome differentiation. The model utilizes the ZY-BERT, a large pre-trained model in the TCM domain, to extract vector representations of TCM texts. A dual-channel network, comprising an improved Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network, is employed to capture both critical local information and global patterns in TCM texts. Additionally, an attention mechanism is introduced to enhance the model’s ability to learn syndrome-related knowledge, integrating syndrome definition knowledge to improve the model’s ability to differentiate complex syndromes. Experiments conducted on a publicly available TCM syndrome differentiation dataset demonstrate that the proposed model achieves an accuracy of 84.01%, representing an 1.75% improvement in accuracy compared to the best baseline model.

Ensemble stacked model for enhanced identification of sentiments from IMDB reviews

The emergence of social media platforms led to the sharing of ideas, thoughts, events, and reviews. The shared views and comments contain people’s sentiments and analysis of these sentiments has emerged as one of the most popular fields of study. Sentiment analysis in the Urdu language is an important research problem similar to other languages, however, it is not investigated very well. On social media platforms like X (Twitter), billions of native Urdu speakers use the Urdu script which makes sentiment analysis in the Urdu language important. In this regard, an ensemble model RRLS is proposed that stacks random forest, recurrent neural network, logistic regression (LR), and support vector machine (SVM). The Internet Movie Database (IMDB) movie reviews and Urdu tweets are examined in this study using Urdu sentiment analysis. The Urdu hack library was used to preprocess the Urdu data, which includes preprocessing operations including normalizing individual letters, merging them, including spaces, etc. concerning punctuation. The problem of accurately encoding Urdu characters and replacing Arabic letters with their Urdu equivalents is fixed by the normalization module. Several models are adopted in this study for extensive evaluation of their accuracy for Urdu sentiment analysis. While the results promising, among machine learning models, the SVM and LR attained an accuracy of 87%, according to performance criteria such as F-measure, accuracy, recall, and precision. The accuracy of the long short-term memory (LSTM) and bidirectional LSTM (BiLSTM) was 84%. The suggested ensemble RRLS model performs better than other learning algorithms and achieves a 90% accuracy rate, outperforming current methods. The use of the synthetic minority oversampling technique (SMOTE) is observed to improve the performance and lead to 92.77% accuracy.

Multi-Strategy Improved Aquila Optimizer Algorithm and Its Application in Railway Freight Volume Prediction

This study proposes a multi-strategy improved Aquila optimizer (MIAO) to address the key limitations of the original Aquila optimizer (AO). First, a phasor operator is introduced to eliminate excessive control parameters in the X2 phase, transforming it into an adaptive parameter-free process. Second, a flow direction operator enhances the X3 phase by improving population diversity and local exploitation. The MIAO algorithm is applied to optimize Long Short-Term Memory (LSTM) hyperparameters, forming the MIAO_LSTM model for monthly railway freight forecasting. Comprehensive evaluations on 15 benchmark functions show MIAO’s superior performance over SOA, PSO, SSA, and AO. Using freight data (2005–2021), MIAO_LSTM achieves lower MAE, MSE, and RMSE compared to traditional LSTM and hybrid models (SSA_LSTM, PSO_LSTM, etc.). Further, Grey Relational Analysis selects high-correlation features (≥0.8) to boost accuracy. The results validate MIAO_LSTM’s effectiveness for practical freight predictions.

Predicting depression and unravelling its heterogeneous influences in middle-aged and older people populations: a machine learning approach

BackgroundAging has become a global trend, and depression, as an accompanying issue, poses a significant threat to the health of middle-aged and older adults. Existing studies primarily rely on statistical methods such as logistic regression for small-scale data analysis, while research on the application of machine learning in large-scale data remains limited. Therefore, this study employs machine learning methods to explore the risk factors for depression among middle-aged and older adults in China.MethodsUsing a two-step hybrid model combining long short-term memory (LSTM) and machine learning (ML), we compared 20 depression risk/protective factors in a balanced panel dataset of middle-aged and elderly Chinese adults (N = 3706; aged 45–94; 64.65% female; 41.20% middle-aged) from the China Health and Retirement Longitudinal Study (CHARLS). Data were collected across five waves (2011, 2013, 2015, 2018, and 2020). The LSTM model predicted risk factors for the fifth wave via data from the preceding four waves. Five ML models were then used to classify depression (yes/no) based on these factors, which included demographic, lifestyle, health, and socioeconomic variables.ResultsThe LSTM model effectively predicted depression-related variables (mean square error = 0.067). The average AUC of the five ML models ranged from 0.78 to 0.82. The key predictive factors were disability, life satisfaction, activities of daily living (ADL) impairment, chronic diseases, and self-reported memory. For the middle-aged group, the top three factors were disability, life satisfaction, and chronic diseases; for the Older people group, they were life satisfaction, chronic diseases, and ADL impairment.ConclusionThe two-step hybrid model ("LSTM + ML") effectively predicted depression over 2 years via demographic and health data, aiding early diagnosis and intervention.

Leveraging Deep Learning for End-to-End Deepfake Video Identification

In recent years, the rise of deepfake technology has posed significant risks to media integrity, cybersecurity, and personal privacy. This project, Leveraging Deep Learning for End-to-End Deepfake Video Identification, proposes a robust detection system integrating spatial and temporal analysis with advanced deep learning techniques. The system employs a hybrid architecture, combining ResNet-50 for detecting frame-level artifacts and Long Short-Term Memory (LSTM) networks for capturing temporal inconsistencies across video frames. The methodology involves preprocessing video data, extracting frames, and applying normalization techniques to enhance feature clarity. ResNet-50 identifies anomalies like irregular facial textures and unnatural lighting, while LSTM tracks facial movements and blink rates. Additional techniques like Optical Flow Analysis, Discrete Fourier Transform (DFT), Principal Component Analysis (PCA), and Recursive Feature Elimination (RFE) further optimize the detection process. Performance evaluation is conducted using benchmark datasets with metrics such as accuracy, precision, recall, and AUC-ROC, demonstrating the system's robustness. A Flask-based web interface allows users to upload videos and obtain real-time feedback. Ethical considerations regarding privacy and responsible use are emphasized, promoting transparency and fairness. In conclusion, this project highlights the potential of combining deep learning and quantum computing paradigms to combat digital content manipulation.

Prediction of pore water pressure generation of liquefiable clean sands under cyclic shear loading through deep learning

ABSTRACT In this study, a novel data-driven approach is carried out to predict the pore pressure generation of liquefiable clean sands during cyclic loading. An extensive and comprehensive database of actual stress-controlled cyclic simple shear test results in terms of pore pressure time histories is gathered from a large number of experiments. While the classical machine learning (ML) algorithms help predict the number of liquefaction cycles in a few models, the desired level of accuracy in predicting the actual trend and robustness in pore pressure build-up is only achieved in deep learning (DL) methods. Results indicate that the Long-Short Term Memory (LSTM) working model, employed with Stacked LSTM and the Windowing data processing method, is necessary for making fairly good cyclic pore pressure build-up predictions. This study proposes a model that can ultimately be utilised to predict the pore pressure response of in-situ liquefiable sandy soil layers without resorting to plasticity-based complex theoretical models, which has been the current practice. The robustness achieved in the model reassures the reliability of the study, raising confidence in developing data-driven constitutive models for soils that have the potential to replace conventional plasticity-based theories.

Surrogate modeling of passive microwave circuits using recurrent neural networks and domain confinement

Electromagnetic (EM) simulation is widespread in microwave engineering. EM tools ensure evaluation reliability but incur significant expenses. These can be mitigated by employing surrogate modeling methods, especially to expedite design workflows like local/global optimization or uncertainty quantification. However, building accurate surrogates is a daunting task beyond simple cases (low dimensionality, narrow geometry parameter and frequency ranges). This research suggests a new technique for dependable modeling of microwave circuits. Its main ingredient is a recurrent neural network (RNN) with the main architectural components being bidirectional Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) layers. These are incorporated to accurately represent frequency relationship within circuit characteristics as well as dependencies between its dimensions and outputs considered as vector-valued functions parameterized by frequency. The network’s hyperparameters are adjusted through Bayesian Optimization (BO). Utilization of frequency as a sequential variable handled by RNN is a distinguishing feature of our approach, which leads to the enhancement of dependability and cost efficiency. Another critical factor is dimension- and volume-wise reduction of the model’s domain achieved through global sensitivity analysis. It allows for additional and dramatic accuracy improvements without diminishing the surrogate’s coverage regarding circuit’s operating parameters. Our methodology has been extensively validated using several microstrip structures. The results demonstrate its competitive performance over a range of kernel-based regression techniques and diverse neural networks. The proposed procedure ensures building models of outstanding predictive power while using small training datasets, which is beyond the capabilities of benchmark algorithms.

Recent Advances in Interactive Driving of Autonomous Vehicles: Comprehensive Review of Approaches

Abstract Interactive autonomous driving is an evolving research domain that demands an autonomous vehicle (AV) to exhibit adaptability to new environments, cognizance of surrounding traffic conditions, and proficient decision-making ability in complex human-dominated scenarios to guarantee safe navigation and promote social compatibility. This paper reviews the diverse methodologies utilized in interactive driving for AVs. Various techniques will be investigated for their unique contributions and capabilities in developing AV systems, such as long short-term memory (LSTM), transformer, artificial potential field (APF), game theory, reinforcement learning (RL)/deep reinforcement learning (DRL), and partially observable Markov decision processes (POMDP), among others. Recent advancements based on these methodologies are summarized to elucidate their application rationale in interactive driving scenarios. The strengths and challenges inherent to each approach within the context of interactive driving are further assessed. Additionally, the resolution of these challenges is explored through integrating different methods. Therefore, a comparative analysis offers crucial perspectives for advancing autonomous driving technologies. This review exclusively focuses on the interactions between AVs and human-driven vehicles (HDVs).

An Integrated Machine Learning and Deep Learning Framework for River-Based Flood Early Warning System

The impact of global climatic conditions and unpredictable rainfall patterns has led to numerous natural disasters in Manipur, including floods and landslides. Developing a reliable flood forecasting and early warning system is crucial to prevent loss of life, property, infrastructure, and crops. This paper presents a machine learning and deep learning framework for flood prediction, comprising four modules: Data Collection, Data Manipulation, Model Implementation, and Early Warning and Flood Forecasting. By leveraging daily data from gauge stations and weather reports, the framework aims to predict river water levels accurately, particularly during the monsoon season. The objectives include developing a model that integrates K-Nearest Neighbor (KNN) with Long Short-Term Memory (LSTM) networks for predicting future water levels using multivariate time series data and providing an algorithm that offers early warnings for potential floods. The proposed framework employs a multivariate time-series approach using a hybrid KNN-LSTM method to predict river water levels. Missing data are imputed using the KNN algorithm with historical weather data, and an LSTM network predicts future water levels based on this information. This approach effectively models temporal dependencies across multiple gauge stations in the Imphal Valley, Manipur. The performance metrics for the proposed model are NSE score of 0.9393, MAPE score of 0.1076, and RMSE score of 1.1828, which are relatively better than those of the conventional approaches. This study presents a novel flood forecasting approach by integrating for imputing missing data with LSTM networks to capture temporal dependencies. The framework is applied to homogeneous regions with correlated time series, effectively modeling the complex dynamics of river water levels across multiple gauge stations.