Risk Prediction Performance Research Articles

Multimodal risk prediction with physiological signals, medical images and clinical notes

The broad adoption of electronic health record (EHR) systems brings us a tremendous amount of clinical data and thus provides opportunities to conduct data-based healthcare research to solve various clinical problems in the medical domain. Machine learning and deep learning methods are widely used in the medical informatics and healthcare domain due to their power to mine insights from raw data. When adapting deep learning models for EHR data, it is essential to consider its heterogeneous nature: EHR contains patient records from various sources including medical tests (e.g. blood test, microbiology test), medical imaging, diagnosis, medications, procedures, clinical notes, etc. Those modalities together provide a holistic view of patient health status and complement each other. Therefore, combining data from multiple modalities that are intrinsically different is challenging but intuitively promising in deep learning for EHR. To assess the expectations of multimodal data, we introduce a comprehensive fusion framework designed to integrate temporal variables, medical images, and clinical notes in EHR for enhanced performance in clinical risk prediction. Early, joint, and late fusion strategies are employed to combine data from various modalities effectively. We test the model with three predictive tasks: in-hospital mortality, long length of stay, and 30-day readmission. Experimental results show that multimodal models outperform uni-modal models in the tasks involved. Additionally, by training models with different input modality combinations, we calculate the Shapley value for each modality to quantify their contribution to multimodal performance. It is shown that temporal variables tend to be more helpful than CXR images and clinical notes in the three explored predictive tasks.

Predicting Risk of Colorectal Cancer After Adenoma Removal in a Large Community-Based Setting.

Colonoscopy surveillance guidelines categorize individuals as high or low risk for future colorectal cancer (CRC) based primarily on their prior polyp characteristics, but this approach is imprecise, and consideration of other risk factors may improve postpolypectomy risk stratification. Among patients who underwent a baseline colonoscopy with removal of a conventional adenoma in 2004-2016, we compared the performance for postpolypectomy CRC risk prediction (through 2020) of a comprehensive model featuring patient age, diabetes diagnosis, and baseline colonoscopy indication and prior polyp findings (i.e., adenoma with advanced histology, polyp size ≥10 mm, and sessile serrated adenoma or traditional serrated adenoma) with a polyp model featuring only polyp findings. Models were developed using Cox regression. Performance was assessed using area under the receiver operating characteristic curve (AUC) and calibration by the Hosmer-Lemeshow goodness-of-fit test. Among 95,001 patients randomly divided 70:30 into model development (n = 66,500) and internal validation cohorts (n = 28,501), 495 CRC were subsequently diagnosed; 354 in the development cohort and 141 in the validation cohort. Models demonstrated adequate calibration, and the comprehensive model demonstrated superior predictive performance to the polyp model in the development cohort (AUC 0.71, 95% confidence interval [CI] 0.68-0.74 vs AUC 0.61, 95% CI 0.58-0.64, respectively) and validation cohort (AUC 0.70, 95% CI 0.65-0.75 vs AUC 0.62, 95% CI 0.57-0.67, respectively). A comprehensive CRC risk prediction model featuring patient age, diabetes diagnosis, and baseline colonoscopy indication and polyp findings was more accurate at predicting postpolypectomy CRC diagnosis than a model based on polyp findings alone.

Incorporating functional annotation with bilevel continuous shrinkage for polygenic risk prediction

BackgroundGenetic variants can contribute differently to trait heritability by their functional categories, and recent studies have shown that incorporating functional annotation can improve the predictive performance of polygenic risk scores (PRSs). In addition, when only a small proportion of variants are causal variants, PRS methods that employ a Bayesian framework with shrinkage can account for such sparsity. It is possible that the annotation group level effect is also sparse. However, the number of PRS methods that incorporate both annotation information and shrinkage on effect sizes is limited. We propose a PRS method, PRSbils, which utilizes the functional annotation information with a bilevel continuous shrinkage prior to accommodate the varying genetic architectures both on the variant-specific level and on the functional annotation level.ResultsWe conducted simulation studies and investigated the predictive performance in settings with different genetic architectures. Results indicated that when there was a relatively large variability of group-wise heritability contribution, the gain in prediction performance from the proposed method was on average 8.0% higher AUC compared to the benchmark method PRS-CS. The proposed method also yielded higher predictive performance compared to PRS-CS in settings with different overlapping patterns of annotation groups and obtained on average 6.4% higher AUC. We applied PRSbils to binary and quantitative traits in three real world data sources (the UK Biobank, the Michigan Genomics Initiative (MGI), and the Korean Genome and Epidemiology Study (KoGES)), and two sources of annotations: ANNOVAR, and pathway information from the Kyoto Encyclopedia of Genes and Genomes (KEGG), and demonstrated that the proposed method holds the potential for improving predictive performance by incorporating functional annotations.ConclusionsBy utilizing a bilevel shrinkage framework, PRSbils enables the incorporation of both overlapping and non-overlapping annotations into PRS construction to improve the performance of genetic risk prediction. The software is available at https://github.com/styvon/PRSbils.

Advances in risk predictive performance of pre-symptomatic type 1 diabetes via the multiplex Antibody-Detection-by-Agglutination-PCR assay.

Achieving early diagnosis of pre-symptomatic type 1 diabetes is critical to reduce potentially life-threatening diabetic ketoacidosis (DKA) at symptom onset, link patients to FDA approved therapeutics that can delay disease progression and support novel interventional drugs development. The presence of two or more islet autoantibodies in pre-symptomatic type 1 diabetes patients indicates high-risk of progression to clinical manifestation. Herein, we characterized the capability of multiplex ADAP assay to predict type 1 diabetes progression. We obtained retrospective coded sera from a cohort of 48 progressors and 44 non-progressors from the NIDDK DPT-1 study. The multiplex ADAP assay and radiobinding assays had positive predictive value (PPV)/negative predictive value (NPV) of 68%/92% and 67%/66% respectively. The improved NPV stemmed from 12 progressors tested positive for multiple islet autoantibodies by multiplex ADAP assay but not by RBA. Furthermore, 6 out of these 12 patients tested positive for multiple islet autoantibodies by RBA in subsequent sampling events with a median delay of 2.8 years compared to multiplex ADAP assay. In summary, multiplex ADAP assay could be an ideal tool for type 1 diabetes risk testing due to its sample-sparing nature (4µL), non-radioactiveness, compatibility with widely available real-time qPCR instruments and favorable risk prediction capability.

Impact of individual level uncertainty of lung cancer polygenic risk score (PRS) on risk stratification.

Although polygenic risk score (PRS) has emerged as a promising tool for predicting cancer risk from genome-wide association studies (GWAS), the individual-level accuracy of lung cancer PRS and the extent to which its impact on subsequent clinical applications remains largely unexplored. Lung cancer PRSs and confidence/credible interval (CI) were constructed using two statistical approaches for each individual: (1) the weighted sum of 16 GWAS-derived significant SNP loci and the CI through the bootstrapping method (PRS-16-CV) and (2) LDpred2 and the CI through posteriors sampling (PRS-Bayes), among 17,166 lung cancer cases and 12,894 controls with European ancestry from the International Lung Cancer Consortium. Individuals were classified into different genetic risk subgroups based on the relationship between their own PRS mean/PRS CI and the population level threshold. Considerable variances in PRS point estimates at the individual level were observed for both methods, with an average standard deviation (s.d.) of 0.12 for PRS-16-CV and a much larger s.d. of 0.88 for PRS-Bayes. Using PRS-16-CV, only 25.0% of individuals with PRS point estimates in the lowest decile of PRS and 16.8% in the highest decile have their entire 95% CI fully contained in the lowest and highest decile, respectively, while PRS-Bayes was unable to find any eligible individuals. Only 19% of the individuals were concordantly identified as having high genetic risk (> 90th percentile) using the two PRS estimators. An increased relative risk of lung cancer comparing the highest PRS percentile to the lowest was observed when taking the CI into account (OR = 2.73, 95% CI: 2.12-3.50, P-value = 4.13 × 10-15) compared to using PRS-16-CV mean (OR = 2.23, 95% CI: 1.99-2.49, P-value = 5.70 × 10-46). Improved risk prediction performance with higher AUC was consistently observed in individuals identified by PRS-16-CV CI, and the best performance was achieved by incorporating age, gender, and detailed smoking pack-years (AUC: 0.73, 95% CI = 0.72-0.74). Lung cancer PRS estimates using different methods have modest correlations at the individual level, highlighting the importance of considering individual-level uncertainty when evaluating the practical utility of PRS.

Penalized deep partially linear cox models with application to CT scans of lung cancer patients.

Lung cancer is a leading cause of cancer mortality globally, highlighting the importance of understanding its mortality risks to design effective patient-centered therapies. The National Lung Screening Trial (NLST) employed computed tomography texture analysis, which provides objective measurements of texture patterns on CT scans, to quantify the mortality risks of lung cancer patients. Partially linear Cox models have gained popularity for survival analysis by dissecting the hazard function into parametric and nonparametric components, allowing for the effective incorporation of both well-established risk factors (such as age and clinical variables) and emerging risk factors (eg, image features) within a unified framework. However, when the dimension of parametric components exceeds the sample size, the task of model fitting becomes formidable, while nonparametric modeling grapples with the curse of dimensionality. We propose a novel Penalized Deep Partially Linear Cox Model (Penalized DPLC), which incorporates the smoothly clipped absolute deviation (SCAD) penalty to select important texture features and employs a deep neural network to estimate the nonparametric component of the model. We prove the convergence and asymptotic properties of the estimator and compare it to other methods through extensive simulation studies, evaluating its performance in risk prediction and feature selection. The proposed method is applied to the NLST study dataset to uncover the effects of key clinical and imaging risk factors on patients' survival. Our findings provide valuable insights into the relationship between these factors and survival outcomes.

Prediction of non emergent acute care utilization and cost among patients receiving Medicaid

Patients receiving Medicaid often experience social risk factors for poor health and limited access to primary care, leading to high utilization of emergency departments and hospitals (acute care) for non-emergent conditions. As programs proactively outreach Medicaid patients to offer primary care, they rely on risk models historically limited by poor-quality data. Following initiatives to improve data quality and collect data on social risk, we tested alternative widely-debated strategies to improve Medicaid risk models. Among a sample of 10 million patients receiving Medicaid from 26 states and Washington DC, the best-performing model tripled the probability of prospectively identifying at-risk patients versus a standard model (sensitivity 11.3% [95% CI 10.5, 12.1%] vs 3.4% [95% CI 3.0, 4.0%]), without increasing “false positives” that reduce efficiency of outreach (specificity 99.8% [95% CI 99.6, 99.9%] vs 99.5% [95% CI 99.4, 99.7%]), and with a ~ tenfold improved coefficient of determination when predicting costs (R2: 0.195–0.412 among population subgroups vs 0.022–0.050). Our best-performing model also reversed the lower sensitivity of risk prediction for Black versus White patients, a bias present in the standard cost-based model. Our results demonstrate a modeling approach to substantially improve risk prediction performance and equity for patients receiving Medicaid.

A polygenic risk score of atrial fibrillation improves prediction of lifetime risk for heart failure.

Heart failure (HF) has shared genetic architecture with its risk factors: atrial fibrillation (AF), body mass index (BMI), coronary heart disease (CHD), systolic blood pressure (SBP), and type 2 diabetes (T2D). We aim to assess the association and risk prediction performance of risk-factor polygenic risk scores (PRSs) for incident HF and its subtypes in bi-racial populations. Five PRSs were constructed for AF, BMI, CHD, SBP, and T2D in White participants of the Atherosclerosis Risk in Communities (ARIC) study. The associations between PRSs and incident HF and its subtypes were assessed using Cox models, and the risk prediction performance of PRSs was assessed using C statistics. Replication was performed in the ARIC study Black and Cardiovascular Health Study (CHS) White participants. In 8624 ARIC study Whites, 1922 (31% cumulative incidence) HF cases developed over 30years of follow-up. PRSs of AF, BMI, and CHD were associated with incident HF (P<0.001), where PRSAF showed the strongest association [hazard ratio (HR): 1.47, 95% confidence interval (CI): 1.41-1.53]. Only the addition of PRSAF to the ARIC study HF risk equation improved C statistics for 10year risk prediction from 0.812 to 0.829 (∆C: 0.017, 95% CI: 0.009-0.026). The PRSAF was associated with both incident HF with reduced ejection fraction (HR: 1.43, 95% CI: 1.27-1.60) and incident HF with preserved ejection fraction (HR: 1.46, 95% CI: 1.33-1.62). The associations between PRSAF and incident HF and its subtypes, as well as the improved risk prediction, were replicated in the ARIC study Blacks and the CHS Whites (P<0.050). Protein analyses revealed that N-terminal pro-brain natriuretic peptide and other 98 proteins were associated with PRSAF. The PRSAF was associated with incident HF and its subtypes and had significant incremental value over an established HF risk prediction equation.

MedDiffusion: Boosting Health Risk Prediction via Diffusion-based Data Augmentation.

Health risk prediction aims to forecast the potential health risks that patients may face using their historical Electronic Health Records (EHR). Although several effective models have developed, data insufficiency is a key issue undermining their effectiveness. Various data generation and augmentation methods have been introduced to mitigate this issue by expanding the size of the training data set through learning underlying data distributions. However, the performance of these methods is often limited due to their task-unrelated design. To address these shortcomings, this paper introduces a novel, end-to-end diffusion-based risk prediction model, named MedDiffusion. It enhances risk prediction performance by creating synthetic patient data during training to enlarge sample space. Furthermore, MedDiffusion discerns hidden relationships between patient visits using a step-wise attention mechanism, enabling the model to automatically retain the most vital information for generating high-quality data. Experimental evaluation on four real-world medical datasets demonstrates that MedDiffusion outperforms 14 cutting-edge baselines in terms of PR-AUC, F1, and Cohen's Kappa. We also conduct ablation studies and benchmark our model against GAN-based alternatives to further validate the rationality and adaptability of our model design. Additionally, we analyze generated data to offer fresh insights into the model's interpretability. The source code is available via https://shorturl.at/aerT0.

Improving supply chain risk assessment with artificial neural network predictions

Operational excellence serves as a cornerstone for the success of businesses,and effective risk management is key for minimizing disruptions, and ensuring business continuity. This paper proposes an innovative methodology that harnesses the power of machine learning in supply chain risk assessment to enhance the ability of organizations to identify, predict, and mitigate various risks that can impact their efficiency, effectiveness, and resilience. This study addresses the inherent subjectivity in human assessment which presents a significant challenges and potential biases in the evaluation process. Auditors, who play a crucial role in identifying and assessing risks within an organization's operations, often rely on subjective judgments influenced by their experiences, expertise, and personal biases. To mitigate this issue, we employ a deconstruction approach, breaking down risk factors into sub-factors, and leverage an Artificial Neural Network model as a predictive tool for accurate risk level predictions and enhanced assessment objectivity. Real-world data from a global automotive company specializing in wiring harnesses are utilized to train the Neural Network model, on a dataset of 2100 samples, exhibits good performance of risk prediction as evaluated by appropriate metrics such as Determination Coefficients and Mean Square Error. Overall, this research contributes to the advancement of risk management practices addressing the challenges of subjectivity in human assessment, to more objective by providing a reliable and data-driven framework that supports managers in strategic decision-making and fortifies supply chain operations through an early risk alarm, empowering organizations to proactively manage risks and achieve autonomy in effective risk management.

A non-socially-sensitive predictive model of prostate cancer for Asian males with benign prostatic hyperplasia: A multi-site cross-sectional case-control study

Background Benign prostatic hyperplasia (BPH) is common in aging Asian males and is associated with an excess risk of developing prostate cancer (PCa). However, discussions about socially-sensitive experiences such as sexual activity, which can significantly predict PCa risk, may be considered stigmatized in Asian culture. This study aimed to develop a predictive model for PCa risk in Asian males with BPH using non-socially-sensitive information. Methods A cross-sectional case-control study, with PCa patients as the cases and remaining as the controls, was conducted on a cohort of Taiwanese males with BPH from four medical institutions. Patients who met the inclusion criteria were enrolled, excluding those aged over 86 years or who had received human papillomavirus (HPV) vaccination. Non-socially-sensitive variables such as obesity, occupational exposure, HPV infection, and PCa family history score (FH score) were included in a fully adjusted logistic regression model, and depicted using a nomogram. Results Among 236 BPH patients, 45.3% had PCa. Obesity, occupational exposure, HPV infection, and family history of PCa were significantly associated with PCa risk. The FH score (OR = 1.89, 95% CI = 1.03–3.47, P = 0.041) had the highest impact, followed by HPV infection (OR = 1.47, 95% CI = 1.03–2.11, P = 0.034), occupational exposure (OR = 1.32, 95% CI = 1.15–1.51, P &lt;0.001), and obesity (OR = 1.22, 95% CI = 1.07–1.41, P = 0.005). The nomogram accurately depicted the predictive risk, and the model demonstrated robust performance compared to individual factors. In addition, the subgroup analysis results showed elderly age group could obtain more favorable predictive performance in our proposed model (AUC = 0.712). Conclusion This non-socially-sensitive predictive model for PCa risk in Taiwanese males with BPH integrates multiple factors that could provide acceptable PCa risk-predictive performance, especially for elderly BPH patients over 70 years, aiding clinical decision-making and early cancer detection.

Discussion on the Enterprise Financial Risk Management Framework Based on AI Fintech

Deep learning algorithms lack interpretability and interpretability in the decision-making process. This makes it difficult to understand the judgment basis and decision-making process of financial risks based on algorithms, which may reduce the trust and acceptance of risk decisions by enterprises. To address this issue, this study introduces the improved random forest algorithm based on the decision tree algorithm to discuss its framework. Through analysis of the PR curve in the experiment, it was determined that the AP value of the enhanced random forest algorithm is 0.9919, a significant improvement over the RF algorithm's previous value of 0.9237. It also has a good balance between the precision rate and the recall rate. By introducing the data set to compare and analyze the three algorithms of SVM, CRAT, and the improved random forest algorithm, it is found that the improved random forest algorithm has a higher test value. Through cluster analysis, it is found that the clustering accuracy of the improved random forest algorithm is about 81%. The final analysis of a company’s financial data sample showed an accuracy rate of 69.6% on the enterprise access index. The improved random forest algorithm achieves a good balance between accuracy and recall, resulting in the potential for high accuracy in the domain of risk assessment. In addition, compared with other algorithms such as SVM and CRAT, the improved random forest algorithm has higher test values, indicating its excellent performance in financial risk prediction. To sum up, this study firstly verified the feasibility and accuracy of the improved random forest algorithm for financial risk prediction. Additionally, it validated the method's predictive capability using factual enterprise data samples and ultimately established a foundation for the enterprise's financial risk management framework. Construction offers a new theoretical direction.

Cardiac murmur grading and risk analysis of cardiac diseases based on adaptable heterogeneous-modality multi-task learning.

Cardiovascular disease (CVDs) has become one of the leading causes of death, posing a significant threat to human life. The development of reliable Artificial Intelligence (AI) assisted diagnosis algorithms for cardiac sounds is of great significance for early detection and treatment of CVDs. However, there is scarce research in this field. Existing research mainly faces three major challenges: (1) They mainly limited to murmur classification and cannot achieve murmur grading, but attempting both classification and grading may lead to negative effects between different multi-tasks. (2) They mostly pay attention to unstructured cardiac sound modality and do not consider the structured demographic modality, as it is difficult to balance the influence of heterogeneous modalities. (3) Deep learning methods lack interpretability, which makes it challenging to apply them clinically. To tackle these challenges, we propose a method for cardiac murmur grading and cardiac risk analysis based on heterogeneous modality adaptive multi-task learning. Specifically, a Hierarchical Multi-Task learning-based cardiac murmur detection and grading method (HMT) is proposed to prevent negative interference between different tasks. In addition, a cardiac risk analysis method based on Heterogeneous Multi-modal feature impact Adaptation (HMA) is also proposed, which transforms unstructured modality into structured modality representation, and utilizes an adaptive mode weight learning mechanism to balance the impact between unstructured modality and structured modality, thus enhancing the performance of cardiac risk prediction. Finally, we propose a multi-task interpretability learning module that incorporates an important evaluation using random masks. This module utilizes SHAP graphs to visualize crucial murmur segments in cardiac sound and employs a multi-factor risk decoupling model based on nomograms. And then we gain insights into the cardiac disease risk in both pre-decoupled multi-modality and post-decoupled single-modality scenarios, thus providing a solid foundation for AI assisted cardiac murmur grading and risk analysis. Experimental results on a large real-world CirCor DigiScope PCG dataset demonstrate that the proposed method outperforms the state-of-the-art (SOTA) method in murmur detection, grading, and cardiac risk analysis, while also providing valuable diagnostic evidence.

Traffic Accident Risk Prediction via Multi-View Multi-Task Spatio-Temporal Networks

Abnormal traffic incidents such as traffic accidents have become a significant health and development threat with the rapid urbanization of many countries. The challenges of accurate traffic risk forecasting are three-fold. First, traffic accident data in some areas of a city is sparse, especially for a fine-grained prediction, which may cause the zero inflation problem during model training. Second, the spatio-temporal correlations of the traffic accidents are rather complex and non-linear, which is difficult to capture by existing shallow models like regression. Third, the occurrence of traffic accidents can be significantly affected by various context features including weather, POI and road network features. To address the above challenges, this paper proposes a Multi-View Multi-Task Spatio-Temporal Networks (MVMT-STN) model to forecast fine- and coarse-grained traffic accident risks of a city simultaneously. Specifically, to address the data sparsity issue in a fine-grained prediction, we adopt a multi-task learning framework to jointly forecast both fine- and coarse-grained traffic accident risks by considering their spatial associations. We conduct extensive experiments over two large real traffic accident datasets. The results show that MVMT-STN improves the performance of traffic accident risk prediction in both fine- and coarse-grained prediction by a large margin compared with existing state-of-the-art.

Enhancing credit risk prediction based on ensemble tree‐based feature transformation and logistic regression

AbstractThe assessment of credit risk for P2P lending platform applicants is critical to investors. Feature engineering is an essential technique in distilling classification knowledge during the credit risk prediction data preprocessing stage. Although previous literature used feature selection methods to identify key features, feature transformation is more useful in discovering intrinsic nonlinear characteristics in credit data. In this study, we propose a synthetic multiple tree‐based feature transformation method to generate features. Multiple tree‐based feature transformation methods are employed and fused to acquire a new feature set. The bagging‐based tree ensemble feature transformation method (Bagging‐TreeEnsembleFT) and boosting‐based tree ensemble feature transformation method (Boosting‐TreeEnsembleFT) are two types of feature transformation methods that we specifically propose to validate their effect. We verify the credit risk prediction performance using the proposed synthetic feature transformation methods on real P2P Lending credit datasets. Empirical analysis demonstrates that tree‐based ensemble feature transformation methods with boosting ensemble strategy achieve better prediction performance on various datasets corresponding to different partitions and class distributions compared to tree‐based ensemble feature transformation methods with bagging ensemble strategy and individuals. Moreover, the proposed synthetic feature transformation method improves the credit risk prediction performance in terms of accuracy, AUC, and F1‐score.

Machine learning-based prediction of acute mortality in emergency department patients using twelve-lead electrocardiogram.

The risk of mortality is relatively high among patients who visit the emergency department (ED), and stratifying patients at high risk can help improve medical care. This study aimed to create a machine-learning model that utilizes the standard 12-lead ECG to forecast acute mortality risk in ED patients. The database included patients who visited the EDs and underwent standard 12-lead ECG between October 2007 and December 2017. A convolutional neural network (CNN) ECG model was developed to classify survival and mortality using 12-lead ECG tracings acquired from 345,593 ED patients. For machine learning model development, the patients were randomly divided into training, validation and testing datasets. The performance of the mortality risk prediction in this model was evaluated for various causes of death. Patients who visited the ED and underwent one or more ECG examinations experienced a high incidence of 30-day mortality [18,734 (5.42%)]. The developed CNN model demonstrated high accuracy in predicting acute mortality (hazard ratio 8.50, 95% confidence interval 8.20-8.80) with areas under the receiver operating characteristic (ROC) curve of 0.84 for the 30-day mortality risk prediction models. This CNN model also demonstrated good performance in predicting one-year mortality (hazard ratio 3.34, 95% confidence interval 3.30-3.39). This model exhibited good predictive performance for 30-day mortality not only for cardiovascular diseases but also across various diseases. The machine learning-based ECG model utilizing CNN screens the risks for 30-day mortality. This model can complement traditional early warning scoring indexes as a useful screening tool for mortality prediction.

The Role of Artificial Intelligence, Financial and Non-Financial Data in Credit Risk Prediction: Literature Review

&#x0D; &#x0D; &#x0D; &#x0D; Small and medium-sized enterprises (SMEs) are of major importance in world economies and job creation. Financing is one of the key issues for SME development since SMEs are often considered riskier than large companies. It is argued in the literature that artificial intelligence (AI) and non-financial data could increase the financial inclusion of disadvantaged groups, such as SMEs. This article presents an overview of selected studies on credit risk prediction from the 1960s to 2022, covering topics of research work applying classical statistical methods, studies using AI methods on traditional financial data and studies applying AI methods on non-financial data. Literature overview results showed that the inclusion of non-financial data in credit risk prediction models could increase credit risk prediction performance, while AI methods can enable the inclusion of non-financial data. Since non-financial data potentially could be used as alternative data in credit prediction models, AI and non-financial data could help to increase access to finance for SMEs&#x0D; &#x0D; &#x0D; &#x0D;

Exploration on Portfolio Selection and Risk Prediction in Financial Markets Based on SVM Algorithm

In order to cope with the complex risk environment of the current financial market, achieve portfolio optimization and accurate risk prediction, this paper conducts effective research using SVM algorithm. This article uses stock data as a sample to empirically analyze the risk return and risk prediction performance of investment portfolio strategies based on SVM algorithm. Compared with traditional index fund investment strategies, the risk resistance of investment portfolio strategies is significantly improved, and the risk return is also stable at a high level. In addition, with the support of SVM algorithm, the risk prediction error level in the financial market remains within a relatively low range. From the perspective of practical applications, the financial market investment portfolio selection and risk prediction based on SVM algorithm has strong feasibility.

Multimodal Machine Learning for 10-Year Dementia Risk Prediction: The Framingham Heart Study.

Early prediction of dementia risk is crucial for effective interventions. Given the known etiologic heterogeneity, machine learning methods leveraging multimodal data, such as clinical manifestations, neuroimaging biomarkers, and well-documented risk factors, could predict dementia more accurately than single modal data. This study aims to develop machine learning models that capitalize on neuropsychological (NP) tests, magnetic resonance imaging (MRI) measures, and clinical risk factors for 10-year dementia prediction. This study included participants from the Framingham Heart Study, and various data modalities such as NP tests, MRI measures, and demographic variables were collected. CatBoost was used with Optuna hyperparameter optimization to create prediction models for 10-year dementia risk using different combinations of data modalities. The contribution of each modality and feature for the prediction task was also quantified using Shapley values. This study included 1,031 participants with normal cognitive status at baseline (age 75±5 years, 55.3% women), of whom 205 were diagnosed with dementia during the 10-year follow-up. The model built on three modalities demonstrated the best dementia prediction performance (AUC 0.90±0.01) compared to single modality models (AUC range: 0.82-0.84). MRI measures contributed most to dementia prediction (mean absolute Shapley value: 3.19), suggesting the necessity of multimodal inputs. This study shows that a multimodal machine learning framework had a superior performance for 10-year dementia risk prediction. The model can be used to increase vigilance for cognitive deterioration and select high-risk individuals for early intervention and risk management.

Risk Prediction of Thalassemia Using Data Mining Classifiers

Medical data mining is concerned with prediction knowledge, which is a useful method for extracting hidden patterns from given data for specific purposes. Thalassemia is one of the most common inherited blood hematological disorders, and this paper adopted data mining classification techniques to generate results with high performance and accuracy for risk prediction of thalassemia. The dataset for this purpose was collected from NIBD (National Institute of Blood Diseases), a well-known institute and hospital for blood diseases in Karachi, Pakistan. They provided 301 records of CBC test reports containing positive and negative statuses of diagnosis of thalassemia traits. There were many instances in the report, of which 6 were used for our research purpose, i.e. Gender, MCV, HGB, HCT, MCHC, and RDW. The dataset was divided into training and test data using the WEKA tool. Four algorithms of data mining classification, namely J48 Decision Tree, Naïve Bayesian Network, SMO algorithm, and Multilayer Perceptron Neural Network were adopted to train the model and classify the patient having traits of thalassemia from normal persons with the use of the WEKA tool. Results revealed that out of all four algorithms, Naïve Bayes provided results with the highest accuracy of 99%.