Multilayer Perceptron Research Articles

Development of Machine Learning-Based Production Forecasting for Offshore Gas Fields Using a Dynamic Material Balance Equation

Offshore oil and gas fields pose significant challenges due to their lower accessibility compared to onshore fields. To enhance operational efficiency in these deep-sea environments, it is essential to design optimal fluid production conditions that ensure equipment durability and flow safety. This study aims to develop a smart operational solution that integrates data from three offshore gas fields with a dynamic material balance equation (DMBE) method. By combining the material balance equation and inflow performance relation (IPR), we establish a reservoir flow analysis model linked to an AI-trained production pipe and subsea pipeline flow analysis model. We simulate time-dependent changes in reservoir production capacity using DMBE and IPR. Additionally, we utilize SLB’s PIPESIM software to create a vertical flow performance (VFP) table under various conditions. Machine learning techniques train this VFP table to analyze pipeline flow characteristics and parameter correlations, ultimately developing a model to predict bottomhole pressure (BHP) for specific production conditions. Our research employs three methods to select the deep learning model, ultimately opting for a multilayer perceptron (MLP) combined with regression. The trained model’s predictions show an average error rate of within 1.5% when compared with existing commercial simulators, demonstrating high accuracy. This research is expected to enable efficient production management and risk forecasting for each well, thus increasing revenue, minimizing operational costs, and contributing to stable plant operations and predictive maintenance of equipment.

Enhanced Hybrid U-Net Framework for Sophisticated Building Automation Extraction Utilizing Decay Matrix

Automatically extracting buildings from remote sensing imagery using deep learning techniques has become essential for various real-world applications. However, mainstream methods often encounter difficulties in accurately extracting and reconstructing fine-grained features due to the heterogeneity and scale variations in building appearances. To address these challenges, we propose LDFormer, an advanced building segmentation model based on linear decay. LDFormer introduces a multi-scale detail fusion bridge (MDFB), which dynamically integrates shallow features to enhance the representation of local details and capture fine-grained local features effectively. To improve global feature extraction, the model incorporates linear decay self-attention (LDSA) and depthwise large separable kernel multi-layer perceptron (DWLSK-MLP) optimizations in the decoder. Specifically, LDSA employs a linear decay matrix within the self-attention mechanism to address long-distance dependency issues, while DWLSK-MLP utilizes step-wise convolutions to achieve a large receptive field. The proposed method has been evaluated on the Massachusetts, Inria, and WHU building datasets, achieving IoU scores of 76.10%, 82.87%, and 91.86%, respectively. LDFormer demonstrates superior performance compared to existing state-of-the-art methods in building segmentation tasks, showcasing its significant potential for building automation extraction.

HRTBDA: a network for post-disaster building damage assessment based on remote sensing images

ABSTRACT Efficient building damage assessment after disasters is vital for emergency response and loss evaluation, but the task is complicated by diverse building structures and complex environments. Traditional methods using Convolutional Neural Networks (CNNs) struggle to capture global contextual features, limiting damage categorization accuracy. To address this, we introduce the High-Resolution Transformer Architecture for Building Damage Assessment (HRTBDA), which enhances multi-scale feature extraction. A Cross-Attention-Based Spatial Fusion (CSF) module is proposed to utilize the attention mechanism, improving the model’s ability to identify detailed associations in damaged buildings. Additionally, we propose a deep convolution network matching optimization strategy that integrates a multilayer perceptron and expands the receptive field, enhancing global feature perception. HRTBDA’s performance was evaluated on two public datasets and compared with five recent frameworks. The model achieved an F1-score of 86.0% in building localization and 78.4% in damage assessment, with a 4.8% improvement in detecting minor damages. These results demonstrate HRTBDA’s potential for improving building damage assessment and highlight its significant advancements over existing methods.

Estimation Network for Multiple Chemical Parameters of Astragalus Leaves Based on Attention Mechanism and Multivariate Hyperspectral Features

In the context of smart agriculture, accurately estimating plant leaf chemical parameters is crucial for optimizing crop management and improving agricultural yield. Hyperspectral imaging, with its ability to capture detailed spectral information across various wavelengths, has emerged as a powerful tool in this regard. However, the complex and high-dimensional nature of hyperspectral data poses significant challenges in extracting meaningful features for precise estimation. To address this challenge, this study proposes an end-to-end estimation network for multiple chemical parameters of Astragalus leaves based on attention mechanism (AM) and multivariate hyperspectral features (AM-MHENet). We leverage HybridSN and multilayer perceptron (MLP) to extract prominent features from the hyperspectral data of Astragalus membranaceus var. mongholicus (AMM) leaves and stems, as well as the surface and deep soil surrounding AMM roots. This methodology allows us to capture the most significant characteristics present in these hyperspectral data with high precision. The AM is subsequently used to assign weights and integrate the hyperspectral features extracted from different parts of the AMM. The MLP is then employed to simultaneously estimate the chlorophyll content (CC) and nitrogen content (NC) of AMM leaves. Compared with estimation networks that utilize only hyperspectral data from AMM leaves as input, our proposed end-to-end AM-MHENet demonstrates superior estimation performance. Specifically, AM-MHENet achieves an R2 of 0.983, an RMSE of 0.73, an MAE of 0.49, and an RPD of 7.63 for the estimation of CC in AMM leaves. For NC estimation, AM-MHENet achieves an R2 value of 0.977, an RMSE of 0.27, an MAE of 0.16, and an RPD of 6.62. These results underscore AM-MHENet’s effectiveness in significantly enhancing the accuracy of both CC and NC estimation in AMM leaves. Moreover, these findings indirectly suggest a strong correlation between the development of AMM leaves and stems, as well as the surface and deep soil surrounding the roots of AMM, and directly highlight the ability of AM to effectively focus on the relevant spectral features within the hyperspectral data. The findings from this study could offer valuable insights into the simultaneous estimation of multiple chemical parameters in plants, thereby making a contribution to the existing body of research in this field.

Advancing Vehicle Security: Deep Learning based Solution for Defending CAN Networks in the Internet of Vehicles

The Internet of Vehicle (IoV) is revolutionizing the automobile sector by allowing vehicles to interact with one another and with roadside infrastructure. The Controller Area Network (CAN) is a vital component of such smart vehicles, allowing communication between various Electronic Control Units (ECUs). However, the CAN protocol's intrinsic lack of security renders it open to a variety of cyber-attacks, posing substantial hazards to both safety and privacy.This research investigates the use of deep learning with multi-layer perceptron to improve the security of CAN networks inside the IoV framework. We discuss current threats to CAN networks, including spoofing, replay, and denial-of-service attacks, and how deep learning may be used to identify and mitigate these threats efficiently. We propose a unique deep learning-based defense mechanism for real-time threat detection. The suggested method is highly effective in identifying and mitigating potential risks, as evidenced by extensive testing on real-world CAN datasets. Based on our findings, the proposed solution has the potential to considerably enhance the security of CAN networks in the Internet of Vehicles, making car communication systems more secure and reliable.

Neural Network-Enabled Multiparametric Impedance Signal Templating for High throughput Single-Cell Deformability Cytometry Under Viscoelastic Extensional Flows.

Cellular biophysical metrics exhibit systematic alterations during processes, such as metastasis and immune cell activation, which can be used to identify and separate live cell subpopulations for targeting drug screening. Image-based biophysical cytometry under extensional flows can accurately quantify cell deformability based on cell shape alterations but needs extensive image reconstruction, which limits its inline utilization to activate cell sorting. Impedance cytometry can measure these cell shape alterations based on electric field screening, while its frequency response offers functional information on cell viability and interior structure, which are difficult to discern by imaging. Furthermore, 1-D temporal impedance signal trains exhibit characteristic shapes that can be rapidly templated in near real-time to extract single-cell biophysical metrics to activate sorting. We present a multilayer perceptron neural network signal templating approach that utilizes raw impedance signals from cells under extensional flow, alongside its training with image metrics from corresponding cells to derive net electrical anisotropy metrics that quantify cell deformability over wide anisotropy ranges and with minimal errors from cell size distributions. Deformability and electrical physiology metrics are applied in conjunction on the same cell for multiparametric classification of live pancreatic cancer cells versus cancer associated fibroblasts using the support vector machine model.

Neural Approach to Coordinate Transformation for LiDAR–Camera Data Fusion in Coastal Observation

The paper presents research related to coastal observation using a camera and LiDAR (Light Detection and Ranging) mounted on an unmanned surface vehicle (USV). Fusion of data from these two sensors can provide wider and more accurate information about shore features, utilizing the synergy effect and combining the advantages of both systems. Fusion is used in autonomous cars and robots, despite many challenges related to spatiotemporal alignment or sensor calibration. Measurements from various sensors with different timestamps have to be aligned, and the measurement systems need to be calibrated to avoid errors related to offsets. When using data from unstable, moving platforms, such as surface vehicles, it is more difficult to match sensors in time and space, and thus, data acquired from different devices will be subject to some misalignment. In this article, we try to overcome these problems by proposing the use of a point matching algorithm for coordinate transformation for data from both systems. The essence of the paper is to verify algorithms based on selected basic neural networks, namely the multilayer perceptron (MLP), the radial basis function network (RBF), and the general regression neural network (GRNN) for the alignment process. They are tested with real recorded data from the USV and verified against numerical methods commonly used for coordinate transformation. The results show that the proposed approach can be an effective solution as an alternative to numerical calculations, due to process improvement. The image data can provide information for identifying characteristic objects, and the obtained accuracies for platform dynamics in the water environment are satisfactory (root mean square error—RMSE—smaller than 1 m in many cases). The networks provided outstanding results for the training set; however, they did not perform as well as expected, in terms of the generalization capability of the model. This leads to the conclusion that processing algorithms cannot overcome the limitations of matching point accuracy. Further research will extend the approach to include information on the position and direction of the vessel.

Development and validation of a machine learning‐based model of ischemic stroke risk in the Chinese elderly hypertensive population

AbstractMachine learning (ML) has made some significant contributions to stroke prevention, but the stability and accuracy of existing models for clinical applications are uncertain. This study develops and validates an interpretable ML model using metabolic and coagulation biomarkers to predict ischemic stroke in elderly hypertensive patients in Northwest China. The prediction model used 453 electronic medical records for the model building (80% as a training set and 20% as a test set) and 132 for external validation. The final seven key features (D‐dimer, cystatin C, homocysteine, hemoglobin A1c, prothrombin time, low‐density lipoprotein C, and triglyceride glucose‐body mass index) were selected by the advanced approach, elastic net, and classical wrapping approaches. The final model, eXtreme gradient boosting, was identified as having superior performance than the other 9 classifers (random forest, Gaussian process, multilayer perceptron, logistic regression, support vector machine, K‐nearest neighbor, decision tree, Gaussian naive bayes, and ensemble model), with area under the receiver‐operating characteristic curves of 0.97 and 0.94 for the test and external validation sets, respectively. The final model demonstrates excellent stability, accuracy, and clinical usefulness through various metrics and decision curve analysis. Additionally, an online human–machine interface application has been developed for clinical practice to help early identification and intervention for ischemic stroke in elderly hypertensive patients.

Analysis of Machine Learning-Based NLOS Signal Identification Algorithm for UWB Indoor Localization Using CIR Waveform Features

Abstract. Ultra-wideband (UWB) technology stands out among numerous indoor positioning techniques due to its high operating frequency, low interception capability, resistance to multipath effects, and strong penetration. The UWB uses the time-of-arrival (TOA) to estimate the distance between the transmitter and receiver anchors in centimeter accuracy. However, in complex indoor positioning environments, obstacles such as walls, glass windows, metal plates, and wooden doors may block and reflect signals, inevitably causing non-line-of-sight (NLOS) errors that significantly affect positioning accuracy. The NLOS signal has lower signal energy due to the reflections. Thus, the channel impulse responses (CIR) from NLOS and LOS are different. To address the NLOS signal identification issue in UWB positioning, we utilize UWB CIR data collected from various positioning scenarios as the data source. CIR waveform input features are provided for the NLOS signal recognition model, and four machine learning models—Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), K-Nearest Neighbors (KNN), and XGBoost—are trained and optimized for NLOS signal recognition. The aim of the study is to analyze the performance of different machine learning algorithms for NLOS signal recognition in UWB indoor localization using these features. Experimental results indicate that machine learning-based NLOS signal recognition algorithms can achieve an accuracy of approximately 77.46%, precision of 80.46%, and an F1 score of 0.81. Among the four models, the XGBoost model demonstrates generally better recognition performance.

IDLB-Pred: identification of disordered lipid binding residues in protein sequences using convolutional neural network

Proteins, nucleic acids, and lipids all interact with intrinsically disordered protein areas. Lipid-binding regions are involved in a variety of biological processes as well as a number of human illnesses. The expanding body of experimental evidence for these interactions and the dearth of techniques to anticipate them from the protein sequence serve as driving forces. Although large-scale laboratory techniques are considered to be essential for equipment for studying binding residues, they are time consuming and costly, making it challenging for researchers to predict lipid binding residues. As a result, computational techniques are being looked at as a different strategy to overcome this difficulty. To predict disordered lipid-binding residues (DLBRs), we proposed iDLB-Pred predictor utilizing benchmark dataset to compute feature through extraction techniques to identify relevant patterns and information. Various classification techniques, including deep learning methods such as Convolutional Neural Networks (CNNs), Deep Neural Networks (DNNs), Multilayer Perceptrons (MLPs), Recurrent Neural Networks (RNNs), Long Short-Term Memory (LSTM) networks, and Gated Recurrent Units (GRUs), were employed for model training. The proposed model, iDLB-Pred, was rigorously validated using metrics such as accuracy, sensitivity, specificity, and Matthew’s correlation coefficient. The results demonstrate the predictor’s exceptional performance, achieving accuracy rates of 81% on an independent dataset and 86% in 10-fold cross-validation.

Semi-supervised comparative learning compensation method for chemical gas sensor drift.

The gradual and unpredictable variation in chemo-sensory signal responses when exposed to the same analyte under identical conditions, commonly referred to as sensor drift, has long been recognized as one of the most serious challenges faced by chemical sensors. The traditional drift compensation method is both labor-intensive and expensive, as it requires frequent collection and labeling of gas samples for recalibration. Introducing a small number of meaningful drift calibration samples can be an attractive strategy to reduce the computational load and improve the performance of the updated classifier. However, under the influence of drift, new challenges arise due to the difference in the distribution of source and target domain data. This paper proposes a novel algorithm framework called semi-supervised contrastive learning drift compensation (SSCLDC). The framework automatically extracts high-level abstract features based on a multilayer perceptron to better represent the structure of the source data. In addition, to address the issue of data distribution differences caused by drift between the source and target domains. We add a small number of reference sample pairs into the training for semi-supervised learning. Combining a contrastive loss function that can represent the matching degree of paired samples effectively overcomes the problem of sensor drift. The Kennard-Stone sequential algorithm is used to select the representative reference sample from the set of candidate reference samples. Experiments conducted on a widely used long-term chemical gas sensor drift dataset demonstrate that the proposed method outperforms several classic drift compensation techniques, highlighting its effectiveness and practical applicability.

Critical comparison of potential machine learning methods for lightning thermal damage assessment of composite laminates

The present study assesses the potential of using machine learning (ML) methods to predict the extent of lightning thermal damage in fiber-reinforced composite laminates using three supervised machine learning (SML) algorithms: (1) linear regression (LR), (2) decision tree (DT)-based, and (3) MLP models. These models were based on the 10 most significant factors that influence the severity of lightning damage, including three current waveform parameters, four material configurations, and three orthogonal electrical conductivities of each composite. All models demonstrated good performance with coefficient of determination (R2) values between 0.84 ~ 0.96. The multilayer perceptron (MLP) regression model trained with the lightning matrix damage dataset showed the most promising results (R2 > 0.94). Additional hyperparameter optimization was performed to improve the prediction performance of the baseline MLP model. The hyperparameter optimization (Adam optimizer, tanh activation function, and three hidden layers with 234 neurons) slightly improved the performance of the baseline MLP model by ~0.02, but achieved faster convergence. This result suggests that the baseline MLP model trained with the lightning matrix damage dataset is sufficiently accurate and robust. This paper highlights that ML-informed regression models can serve as an efficient first pass-estimator of lightning matrix damage in composite laminates, potentially reducing the amount of extremely time-consuming and expensive laboratory-scale lightning tests or streamlining the development of complex lightning damage models for future design.

Evaluation of asphalt mixtures modified with low-density polyethylene and high-density polyethylene using experimental results and machine learning models

The widespread use of low-density polyethylene (LDPE) and high-density polyethylene (HDPE) plastics has resulted in a large amount of waste plastic that requires appropriate disposal or reuse. One potential solution is to use them in the modification of asphalt concrete (AC) mixtures for more sustainable highways. To study this possibility, permanent deformation and dynamic modulus (DM) of the LDPE and HDPE modified AC mixtures was investigated by conducting flow number (FN), flow time (FT) and DM tests on Superpave gyratory compacted specimens. Machine learning models; multi-layer perceptron (MLP), radial basis function neural network (RBFNN), generalized regression neural network (GRNN) and support vector machine (SVM) were used to predict the DM on the basis of frequency and temperature parameters. The model’s performance was gauged by analyzing the root mean square error, mean relative error, and coefficient of determination. The study findings revealed that the LDPE and HDPE modified AC mixtures provide 2.07 times and 1.27 times better resistance to permanent deformation, respectively, than their counterpart. It was also found that the LDPE and HDPE modified AC mixtures have 2.1 times and 1.4 times higher DM values, respectively, than the Control AC mixtures. Among the machine learning models, MLP (R2 = 0.98) showed best accuracy in predicting DM and thus is recommended to be used in similar studies due to its robustness. Additionally, the feature importance analysis revealed that frequency has the highest impact on DM predictions, followed by temperature and the inclusion of the LDPE.

AI-Powered Approaches for Hypersurface Reconstruction in Multidimensional Spaces

The present article explores the possibilities of using artificial neural networks to solve problems related to reconstructing complex geometric surfaces in Euclidean and pseudo-Euclidean spaces, examining various approaches and techniques for training the networks. The main focus is on the possibility of training a set of neural networks with information about the available surface points, which can then be used to predict and complete missing parts. A method is proposed for using separate neural networks that reconstruct surfaces in different spatial directions, employing various types of architectures, such as multilayer perceptrons, recursive networks, and feedforward networks. Experimental results show that artificial neural networks can successfully approximate both smooth surfaces and those containing singular points. The article presents the results with the smallest error, showcasing networks of different types, along with a technique for reconstructing geographic relief. A comparison is made between the results achieved by neural networks and those obtained using traditional surface approximation methods such as Bézier curves, k-nearest neighbors, principal component analysis, Markov random fields, conditional random fields, and convolutional neural networks.

AI-DPAPT: a machine learning framework for predicting PROTAC activity.

Proteolysis Targeting Chimeras are part of targeted protein degradation (TPD) techniques, which are significant for pharmacological and therapy development. Small-molecule interaction with the targeted protein is a complicated endeavor and a challenge to predict the proteins accurately. This study used machine learning algorithms and molecular fingerprinting techniques to build an AI-powered PROTAC Activity Prediction Tool that could predict PROTAC activity by examining chemical structures. The chemical structures of a diverse set of PROTAC drugs and their corresponding activities are selected as a dataset for training the tool. The processes used in this study included data preparation, feature extraction, and model training. Further, evaluation was done for the performance of the various classifiers, such as AdaBoost, Support Vector Machine, Random Forest, Gradient Boosting, and Multi-Layer Perceptron. The findings show that the methods selected here depict accurate PROTAC activities. All the models in this study showed an ROC curve better than 0.9, while the random forest on the test set of the AI-DPAPT had an area under the curve score of 0.97, thus showing accurate results. Furthermore, the study revealed significant insights into the molecular features that can influence the functions of the PROTAC. These findings can potentially increase the understanding of the structure-activity correlations involved in the TPD. Overall, the investigation contributes to computational drug development by introducing this platform powered by artificial intelligence that predicts the function of PROTAC. In addition, it sped up the processes of identifying and improving previously unknown medications. The AI-DPAPT platform can be accessed online using a web server at https://ai-protac.streamlit.app/ .

Impact of metadata in multimodal classification of bone tumours

The accurate classification of bone tumours is crucial for guiding clinical decisions regarding treatment and follow-up. However, differentiating between various tumour types is challenging due to the rarity of certain entities, high intra-class variability, and limited training data in clinical practice. This study proposes a multimodal deep learning model that integrates clinical metadata and X-ray imaging to improve the classification of primary bone tumours. The dataset comprises 1,785 radiographs from 804 patients collected between 2000 and 2020, including metadata such as age, affected bone site, tumour position, and gender. Ten tumour types were selected, with histopathology or tumour board decisions serving as the reference standard.MethodsOur model is based on the NesT image classification model and a multilayer perceptron with a joint fusion architecture. Descriptive statistics included incidence and percentage ratios for discrete parameters, and mean, standard deviation, median, and interquartile range for continuous parameters.ResultsThe mean age of the patients was 33.62 ± 18.60 years, with 54.73% being male. Our multimodal deep learning model achieved 69.7% accuracy in classifying primary bone tumours, outperforming the Vision Transformer model by five percentage points. SHAP values indicated that age had the most substantial influence among the considered metadata.ConclusionThe joint fusion approach developed in this study, integrating clinical metadata and imaging data, outperformed state-of-the-art models in classifying primary bone tumours. The use of SHAP values provided insights into the impact of different metadata on the model’s performance, highlighting the significant role of age. This approach has potential implications for improving diagnostic accuracy and understanding the influence of clinical factors in tumour classification.

Identification of non-circular profiles in hydrodynamic journal bearings

Hydrodynamic bearings, especially cylindrical radial plain journal bearings, are widely utilized in industry for their high load capacity and low friction energy losses. However, these bearings are prone to faults such as wear and ovalization, which can deform their circular profile and affect their vibrational response. Detecting these faults is essential to reduce their impact on production. This study introduces a methodology to identify hydrodynamic bearings with non-circular profiles. The bearing model and its numerical solution are implemented using the Finite Volume Method, with the effects of failures incorporated into a rotating system modeled by the Finite Element Method. A dataset is generated to reflect three common failure conditions in industrial applications: wear, ovalization, and a combination of ovalization with wear. The authors used this dataset to train a Multilayer Perceptron (MLP) neural network, which can identify the bearing profile shape based on specific attributes of the dynamic responses. The identification tests for the three fault conditions demonstrated high accuracy, particularly in distinguishing between ovalization and wear.

A hybrid predictive modeling approach for catalyzed polymerization reactors

Polymerization reactions are characterized by complex, nonlinear behaviors that pose significant challenges for conventional modeling techniques. Accurate and reliable models are crucial for advancing material science and enabling technological innovations across various industries. Conventional first-principles models often fall short in capturing the intricate dynamics of polymeric systems, leading to limitations in predictive accuracy. In this work, we propose a novel hybrid modeling approach that combines a conventional first-principles model with the strengths of a data-driven multi-layer perceptron (MLP) model and also a linear regression (LR) model to enhance the predictability of polymerization processes. Utilizing this hybrid approach significantly reduces the mean absolute error for predicting the concentrations of main reagents by 84% and 86%, respectively, in experiments with significantly deviant outcomes. Our results indicate that the model is capable of predicting the concentrations of both the main and side products with a maximum error margin of 3.5%.

A Portable Electronic Nose Coupled with Deep Learning for Enhanced Detection and Differentiation of Local Thai Craft Spirits

In this study, our primary focus is the biomimetic design and rigorous evaluation of an economically viable and portable ‘e-nose’ system, tailored for the precise detection of a broad range of volatile organic compounds (VOCs) in local Thai craft spirits. This e-nose system is innovatively equipped with cost-efficient metal oxide gas sensors and a temperature/humidity sensor, ensuring comprehensive and accurate sensing. A custom-designed real-time data acquisition system is integrated, featuring gas flow control, humidity filters, dual sensing/reference chambers, an analog-to-digital converter, and seamless data integration with a laptop. Deep learning, utilizing a multilayer perceptron (MLP), is employed to achieve highly effective classification of local Thai craft spirits, demonstrated by a perfect classification accuracy of 100% in experimental studies. This work underscores the significant potential of biomimetic principles in advancing cost-effective, portable, and analytically precise e-nose systems, offering valuable insights into future applications of advanced gas sensor technology in food, biomedical, and environmental monitoring and safety.

Enhancing indoor PM2.5 predictions based on land use and indoor environmental factors by applying machine learning and spatial modeling approaches

The presence of fine particulate matter (PM2.5) indoors constitutes a significant component of overall PM2.5 exposure, as individuals spend 90% of their time indoors; however, personal monitoring for large cohorts is often impractical. In light of this, this study seeks to employ a novel geospatial artificial intelligence (Geo-AI) coupled with machine learning (ML) approaches to develop indoor PM2.5 models. Multiple predictor variables were collected from 102 residential households, including meteorological data; elevation; land use; indoor environmental factors including human activities, building characteristics, infiltration factors, and real-time measurements; and various other factors. Geo-AI, which integrates land use regression, inverse distance weighting, and ML algorithms, was utilized to construct outdoor PM2.5 and PM10 estimates for residential households. The most influential variables were identified via correlation analysis and stepwise regression. Three ML methods, namely support vector machine, multiple linear regression, and multilayer perceptron (MLP) were used to estimate indoor PM2.5 concentration. Then, MLP was employed to blend three ML algorithms. The resulting model demonstrated commendable performance, achieving a 10-fold cross-validation R2 of 0.92 and a root mean square error of 2.3 μg/m3 for indoor PM2.5 estimations. Notably, the combination of Geo-AI and ensembled ML models in this study outperformed all other individual models. In addition, the present study pointed out the most influential factors for indoor PM2.5 model were outdoor PM2.5, PM2.5/PM10 ratio, sampling month, infiltration factor, located near factory, cleaning frequency, number of door entrance linked with outdoor, and wall material. Further exploration of diverse ensemble model formats to integrate estimates from different models could enhance overall performance. Consequently, the potential applications of this model extend to estimating real individual exposure to PM2.5 for further epidemiological research. Moreover, the model offers valuable insights for efficient indoor air quality management and control strategies.