Deep Learning Predictions Research Articles

The synchronized dynamic release behavior of microplastics during farmland soil erosion process

Microplastics (MPs) are widespread in farmland soil. However, the risks associated with their loss through soil erosion remain unknown. This study investigates the occurrence and behavior of MPs in farmland soil in a southeastern coastal area of China, focusing on their synchronized dynamic release during soil erosion scenarios. The results showed that the abundance of MPs in the tested farmland soil ranged from 2.40×104 to 1.04×105 items·kg-1. MPs predominantly appear as fragments and particles, with sizes concentrated between 30-100 μm. During the process of soil erosion, characterized by rapid soil subsidence, the amount of MPs released into water bodies initially decreases, averaging a reduction of 1.08×104 items·kg-1. This is followed by an average increase of 1.89×104 items·kg-1. The competition between the adsorption, collision, and sedimentation of soil particles and the desorption and release of settled particles, determines this behavior. This pattern is strongly related to the physicochemical properties and mechanical composition of the soil. Deep learning predictions revealed that, without external influences, 49.42% of MPs in farmland soil could be synchronously released into water bodies during erosion. The analysis shows that MPs exhibit dynamic behavior in time and space, posing serious threats to aquatic ecosystems. Controlling soil erosion in farmland is crucial for the source management of MP migration.

Using deep-learning predictions reveals a large number of register errors in PDB depositions.

The accuracy of the information in the Protein Data Bank (PDB) is of great importance for the myriad downstream applications that make use of protein structural information. Despite best efforts, the occasional introduction of errors is inevitable, especially where the experimental data are of limited resolution. A novel protein structure validation approach based on spotting inconsistencies between the residue contacts and distances observed in a structural model and those computationally predicted by methods such as AlphaFold2 has previously been established. It is particularly well suited to the detection of register errors. Importantly, this new approach is orthogonal to traditional methods based on stereochemistry or map-model agreement, and is resolution independent. Here, thousands of likely register errors are identified by scanning 3-5 Å resolution structures in the PDB. Unlike most methods, the application of this approach yields suggested corrections to the register of affected regions, which it is shown, even by limited implementation, lead to improved refinement statistics in the vast majority of cases. A few limitations and confounding factors such as fold-switching proteins are characterized, but this approach is expected to have broad application in spotting potential issues in current accessions and, through its implementation and distribution in CCP4, helping to ensure the accuracy of future depositions.

A Novel Stock Price Prediction and Trading Methodology Based on Active Learning Surrogated with CycleGAN and Deep Learning and System Engineering Integration: A Case Study on TSMC Stock Data

Technical analysis, reliant on statistics and charting tools, is a predominant method for predicting stock prices. However, given the impact of the joint effect of stock price and trading volume, analyses focusing solely on single factors at isolated time points often yield partial or inaccurate results. This study introduces the application of Cycle Generative Adversarial Network (CycleGAN) alongside Deep Learning (DL) models, such as Residual Neural Network (ResNet) and Long Short-Term Memory (LSTM), to assess the joint effects of stock price and trading volume on prediction accuracy. By incorporating these models into system engineering (SE), the research aims to decode short-term stock market trends and improve investment decisions through the integration of predicted stock prices with Bollinger Bands. Thereby, active learning (AL) is employed to avoid over-and under-fitting and find the hyperparameters for the overall system model. Focusing on TSMC’s stock price prediction, the use of CycleGAN for analyzing 30-day stock data showcases the capability of ResNet and LSTM models in achieving high accuracy and F-1 scores for a five-day prediction period. Further analysis reveals that combining DL predictions with SE principles leads to more precise short-term forecasts. Additionally, integrating these predictions with Bollinger Bands demonstrates a decrease in trading frequency and a significant 30% increase in average Return on Investment (ROI). This innovative approach marks a first in the field of stock market prediction, offering a comprehensive framework for enhancing predictive accuracy and investment outcomes.

Sparsity and mixing effects in deep learning predictions of temperature and humidity

Developing deep learning models for predicting environmental data is a powerful tool that can significantly enhance equipment design, optimize the implementation of engineering systems, and deepen our understanding of the limitations imposed by flow physics. This study unequivocally demonstrates the accuracy of forecasting models based on popular deep learning algorithms, such as the long-short-term memory model, in turbulent mixing regions associated with flow physics arising from ventilation. This accuracy is contingent on two essential conditions. First, the sparsity of the sampling data is consistent with the model's accuracy overall. Second, the data sparsity ensures reasonable accuracy in the turbulent mixing regions. The investigation combines high-resolution flow simulation data with deep learning predictions of velocity, temperature, and relative humidity in a ventilated confined space. The results of this study, with their high accuracy, not only help to understand the mixing arising from flow circulation but also pave the way for developing predictive capabilities for environmental data.

Three-Dimensional Deep Learning Normal Tissue Complication Probability Model to Predict Late Xerostomia in Patients With Head and Neck Cancer

Conventional normal tissue complication probability (NTCP) models for patients with head and neck cancer are typically based on single-value variables, which, for radiation-induced xerostomia, are baseline xerostomia and mean salivary gland doses. This study aimed to improve the prediction of late xerostomia by using 3-dimensional information from radiation dose distributions, computed tomography imaging, organ-at-risk segmentations, and clinical variables with deep learning (DL). An international cohort of 1208 patients with head and neck cancer from 2 institutes was used to train and twice validate DL models (deep convolutional neural network, EfficientNet-v2, and ResNet) with 3-dimensional dose distribution, computed tomography scan, organ-at-risk segmentations, baseline xerostomia score, sex, and age as input. The NTCP endpoint was moderate-to-severe xerostomia 12 months postradiation therapy. The DL models' prediction performance was compared with a reference model: a recently published xerostomia NTCP model that used baseline xerostomia score and mean salivary gland doses as input. Attention maps were created to visualize the focus regions of the DL predictions. Transfer learning was conducted to improve the DL model performance on the external validation set. All DL-based NTCP models showed better performance (area under the receiver operating characteristic curve [AUC]test, 0.78-0.79) than the reference NTCP model (AUCtest, 0.74) in the independent test. Attention maps showed that the DL model focused on the major salivary glands, particularly the stem cell-rich region of the parotid glands. DL models obtained lower external validation performance (AUCexternal, 0.63) than the reference model (AUCexternal, 0.66). After transfer learning on a small external subset, the DL model (AUCtl, external, 0.66) performed better than the reference model (AUCtl, external, 0.64). DL-based NTCP models performed better than the reference model when validated in data from the same institute. Improved performance in the external data set was achieved with transfer learning, demonstrating the need for multicenter training data to realize generalizable DL-based NTCP models.

Deep learning guided design of dynamic proteins.

Deep learning has greatly advanced design of highly stable static protein structures, but the controlled conformational dynamics that are hallmarks of natural switch-like signaling proteins have remained inaccessible to de novo design. Here, we describe a general deep-learning-guided approach for de novo design of dynamic changes between intra-domain geometries of proteins, similar to switch mechanisms prevalent in nature, with atom-level precision. We solve 4 structures validating the designed conformations, show microsecond transitions between them, and demonstrate that the conformational landscape can be modulated by orthosteric ligands and allosteric mutations. Physics-based simulations are in remarkable agreement with deep-learning predictions and experimental data, reveal distinct state-dependent residue interaction networks, and predict mutations that tune the designed conformational landscape. Our approach demonstrates that new modes of motion can now be realized through de novo design and provides a framework for constructing biology-inspired, tunable and controllable protein signaling behavior de novo.

Text summarization based on semantic graphs: an abstract meaning representation graph-to-text deep learning approach

Nowadays, due to the constantly growing amount of textual information, automatic text summarization constitutes an important research area in natural language processing. In this work, we present a novel framework that combines semantic graph representations along with deep learning predictions to generate abstractive summaries of single documents, in an effort to utilize a semantic representation of the unstructured textual content in a machine-readable, structured, and concise manner. The overall framework is based on a well defined methodology for performing semantic graph parsing, graph construction, graph transformations for machine learning models and deep learning predictions. The employed semantic graph representation focuses on using the model of abstract meaning representation. Several combinations of graph construction and graph transformation methods are investigated to specify the most efficient of them for the machine learning models. Additionally, a range of deep learning architectures is examined, including a sequence-to-sequence attentive network, reinforcement learning, transformer-based architectures, and pre-trained neural language models. In this direction, a semantic graph representation of an original text is extracted, and then the present framework formulates the problem as a graph-to-summary learning problem to predict a summary of an original text. To the best of our knowledge, this formulation of graph-to-summary predictions in abstractive text summarization, without other intermediate steps in the machine learning phase, has not been presented in the relevant literature. Another important contribution is an introduction of a measure for assessing the factual consistency of the generated summaries in an effort to provide a qualitative evaluation. To assess the framework, an extensive experimental procedure is presented that uses popular datasets to evaluate key aspects of the proposed approach. The obtained results exhibit promising performance, validating the robustness of the proposed framework.

Prostate cancer risk assessment and avoidance of prostate biopsies using fully automatic deep learning in prostate MRI: comparison to PI-RADS and integration with clinical data in nomograms.

Risk calculators (RCs) improve patient selection for prostate biopsy with clinical/demographic information, recently with prostate MRI using the prostate imaging reporting and data system (PI-RADS). Fully-automated deep learning (DL) analyzes MRI data independently, and has been shown to be on par with clinical radiologists, but has yet to be incorporated into RCs. The goal of this study is to re-assess the diagnostic quality of RCs, the impact of replacing PI-RADS with DL predictions, and potential performance gains by adding DL besides PI-RADS. One thousand six hundred twenty-seven consecutive examinations from 2014 to 2021 were included in this retrospective single-center study, including 517 exams withheld for RC testing. Board-certified radiologists assessed PI-RADS during clinical routine, then systematic and MRI/Ultrasound-fusion biopsies provided histopathological ground truth for significant prostate cancer (sPC). nnUNet-based DL ensembles were trained on biparametric MRI predicting the presence of sPC lesions (UNet-probability) and a PI-RADS-analogous five-point scale (UNet-Likert). Previously published RCs were validated as is; with PI-RADS substituted by UNet-Likert (UNet-Likert-substituted RC); and with both UNet-probability and PI-RADS (UNet-probability-extended RC). Together with a newly fitted RC using clinical data, PI-RADS and UNet-probability, existing RCs were compared by receiver-operating characteristics, calibration, and decision-curve analysis. Diagnostic performance remained stable for UNet-Likert-substituted RCs. DL contained complementary diagnostic information to PI-RADS. The newly-fitted RC spared 49% [252/517] of biopsies while maintaining the negative predictive value (94%), compared to PI-RADS ≥ 4 cut-off which spared 37% [190/517] (p < 0.001). Incorporating DL as an independent diagnostic marker for RCs can improve patient stratification before biopsy, as there is complementary information in DL features and clinical PI-RADS assessment. For patients with positive prostate screening results, a comprehensive diagnostic workup, including prostate MRI, DL analysis, and individual classification using nomograms can identify patients with minimal prostate cancer risk, as they benefit less from the more invasive biopsy procedure. The current MRI-based nomograms result in many negative prostate biopsies. The addition of DL to nomograms with clinical data and PI-RADS improves patient stratification before biopsy. Fully automatic DL can be substituted for PI-RADS without sacrificing the quality of nomogram predictions. Prostate nomograms show cancer detection ability comparable to previous validation studies while being suitable for the addition of DL analysis.

In silico design of DNA sequences for in vivo nucleosome positioning.

The computational design of synthetic DNA sequences with designer in vivo properties is gaining traction in the field of synthetic genomics. We propose here a computational method which combines a kinetic Monte Carlo framework with a deep mutational screening based on deep learning predictions. We apply our method to build regular nucleosome arrays with tailored nucleosomal repeat lengths (NRL) in yeast. Our design was validated in vivo by successfully engineering and integrating thousands of kilobases long tandem arrays of computationally optimized sequences which could accommodate NRLs much larger than the yeast natural NRL (namely 197 and 237bp, compared to the natural NRL of ∼165bp). RNA-seq results show that transcription of the arrays can occur but is not driven by the NRL. The computational method proposed here delineates the key sequence rules for nucleosome positioning in yeast and should be easily applicable to other sequence properties and other genomes.

Fairness for Deep Learning Predictions Using Bias Parity Score Based Loss Function Regularization

Rising acceptance of machine learning driven decision support systems underscores the need for ensuring fairness for all stakeholders. This work proposes a novel approach to increase a Neural Network model’s fairness during the training phase. We offer a frame-work to create a family of diverse fairness enhancing regularization components that can be used in tandem with the widely accepted binary-cross-entropy based accuracy loss. We use Bias Parity Score (BPS), a metric that quantifies model bias with a single value, to build loss functions pertaining to different statistical measures — even for those that may not be developed yet. We analyze behavior and impact of the newly minted regularization components on bias. We explore their impact in the realm of recidivism and census-based adult income prediction. The results illustrate that apt fairness loss functions can mitigate bias without forsaking accuracy even for imbalanced datasets.

Deep learning predictions of TCR-epitope interactions reveal epitope-specific chains in dual alpha T cells

T cells have the ability to eliminate infected and cancer cells and play an essential role in cancer immunotherapy. T cell activation is elicited by the binding of the T cell receptor (TCR) to epitopes displayed on MHC molecules, and the TCR specificity is determined by the sequence of its α and β chains. Here, we collect and curate a dataset of 17,715 αβTCRs interacting with dozens of class I and class II epitopes. We use this curated data to develop MixTCRpred, an epitope-specific TCR-epitope interaction predictor. MixTCRpred accurately predicts TCRs recognizing several viral and cancer epitopes. MixTCRpred further provides a useful quality control tool for multiplexed single-cell TCR sequencing assays of epitope-specific T cells and pinpoints a substantial fraction of putative contaminants in public databases. Analysis of epitope-specific dual α T cells demonstrates that MixTCRpred can identify α chains mediating epitope recognition. Applying MixTCRpred to TCR repertoires from COVID-19 patients reveals enrichment of clonotypes predicted to bind an immunodominant SARS-CoV-2 epitope. Overall, MixTCRpred provides a robust tool to predict TCRs interacting with specific epitopes and interpret TCR-sequencing data from both bulk and epitope-specific T cells.

Explainable deep learning model for predicting money laundering transactions

Abstract Money laundering has been a global issue for decades. The ever-changing technology landscape, digital channels, and regulations make it increasingly difficult. Financial institutions use rule-based systems to detect suspicious money laundering transactions. However, it suffers from large false positives (FPs) that lead to operational efforts or misses on true positives (TPs) that increase the compliance risk. This paper presents a study of convolutional neural network (CNN) to predict money laundering and employs SHapley Additive exPlanations (SHAP) explainable artificial intelligence (AI) method to explain the CNN predictions. The results highlight the role of CNN in detecting suspicious transactions with high accuracy and SHAP’s role in bringing out the rationale of deep learning predictions.

Deep Learning-Enabled Holistic Control and Prediction System for Building Energy Consumption and Distribution Optimization

Managing a building’s energy consumption while minimizing costs and grid reliance is complex. A novel system is proposed, blending advanced technologies and methodologies. Deep learning techniques, particularly the fusion of Hough Transform within a Gated Recurrent Unit, form the core of this approach. User-defined preferences are prioritized, offering personalized energy-saving strategies. Simulation and experimentation demonstrate the system’s efficacy, forecasting energy demand and supply with 4%–10% deviation accuracy in hourly data for months ahead. By leveraging deep learning predictions, efficient energy storage, and an optimized scheduling algorithm, the system achieves an 84% reduction in grid energy reliance, cutting power expenses by 87%. Comparative cost analysis over thirty years highlights the system’s cost-effectiveness and the inefficiency of installing wind turbines on the building’s roof. In essence, this system offers a data-driven solution that optimizes energy management, reduces grid dependency, and tailors energy-saving strategies, significantly cutting electricity expenses.

Hidden Markov guided Deep Learning models for forecasting highly volatile agricultural commodity prices

Predicting agricultural commodity prices accurately is of utmost importance due to various factors such as perishability, seasonality, production uncertainty etc. Moreover, the substantial volatility that may be exhibited in time series further adds to the complexity and constitutes a significant challenge. In this paper, a Hidden Markov (HM) guided Deep Learning (DL) models has been developed on nonlinear and nonstationary price data of agricultural commodities for forecasting by considering technical indicators viz., Moving Average (MA), Bollinger Bands (BB), Moving Average Convergence Divergence (MACD), Exponential MA (EMA) and Fast Fourier Transformation (FFT). HM Models (HMMs) can effectively handle the sequential dependencies and hidden states, while DL approach can learn complex patterns and relationships within the price series and thus the drawback of lack of generalization capability in the DL model has been overcome by HMM. In this study, the Potato price data of the Champadanga district of West Bengal, India has been utilized to assess the performance of the proposed technique. HMM has been combined with six baseline DL models viz., Recurrent Neural Networks (RNN), Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Bidirectional LSTM (BiLSTM) and Bidirectional GRU (BiGRU) for forecast modeling. Performance evaluation metrics viz., Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE) and the insightful Diebold–Mariano (DM) test revealed that Hidden Markov hybridized with DL models surpassed baseline DL models in forecasting accuracy for 1-week, 4-week, 8-week and 12-week ahead DL predictions. The proposed approach holds significant promise for enhancing the precision of agricultural commodity price forecasting with far-reaching implications for various stakeholders such as farmers and planners.

Artificial Intelligence Plant Doctor: Plant Disease Diagnosis Using GPT4-vision

Integrated pest management is essential for controlling plant diseases that reduce crop yields. Rapid diagnosis is crucial for effective management in the event of an outbreak to identify the cause and minimize damage. Diagnosis methods range from indirect visual observation, which can be subjective and inaccurate, to machine learning and deep learning predictions that may suffer from biased data. Direct molecular-based methods, while accurate, are complex and time-consuming. However, the development of large multimodal models, like GPT-4, combines image recognition with natural language processing for more accurate diagnostic information. This study introduces GPT-4-based system for diagnosing plant diseases utilizing a detailed knowledge base with 1,420 host plants, 2,462 pathogens, and 37,467 pesticide instances from the official plant disease and pesticide registries of Korea. The AI plant doctor offers interactive advice on diagnosis, control methods, and pesticide use for diseases in Korea and is accessible at https://pdoc.scnu.ac.kr/.

Development and validation of an interpretable model integrating multimodal information for improving ovarian cancer diagnosis

Ovarian cancer, a group of heterogeneous diseases, presents with extensive characteristics with the highest mortality among gynecological malignancies. Accurate and early diagnosis of ovarian cancer is of great significance. Here, we present OvcaFinder, an interpretable model constructed from ultrasound images-based deep learning (DL) predictions, Ovarian–Adnexal Reporting and Data System scores from radiologists, and routine clinical variables. OvcaFinder outperforms the clinical model and the DL model with area under the curves (AUCs) of 0.978, and 0.947 in the internal and external test datasets, respectively. OvcaFinder assistance led to improved AUCs of radiologists and inter-reader agreement. The average AUCs were improved from 0.927 to 0.977 and from 0.904 to 0.941, and the false positive rates were decreased by 13.4% and 8.3% in the internal and external test datasets, respectively. This highlights the potential of OvcaFinder to improve the diagnostic accuracy, and consistency of radiologists in identifying ovarian cancer.

Predicting the Structure of Enzymes with Metal Cofactors: The Example of [FeFe] Hydrogenases.

The advent of deep learning algorithms for protein folding opened a new era in the ability of predicting and optimizing the function of proteins once the sequence is known. The task is more intricate when cofactors like metal ions or small ligands are essential to functioning. In this case, the combined use of traditional simulation methods based on interatomic force fields and deep learning predictions is mandatory. We use the example of [FeFe] hydrogenases, enzymes of unicellular algae promising for biotechnology applications to illustrate this situation. [FeFe] hydrogenase is an iron-sulfur protein that catalyzes the chemical reduction of protons dissolved in liquid water into molecular hydrogen as a gas. Hydrogen production efficiency and cell sensitivity to dioxygen are important parameters to optimize the industrial applications of biological hydrogen production. Both parameters are related to the organization of iron-sulfur clusters within protein domains. In this work, we propose possible three-dimensional structures of Chlorella vulgaris 211/11P [FeFe] hydrogenase, the sequence of which was extracted from the recently published genome of the given strain. Initial structural models are built using: (i) the deep learning algorithm AlphaFold; (ii) the homology modeling server SwissModel; (iii) a manual construction based on the best known bacterial crystal structure. Missing iron-sulfur clusters are included and microsecond-long molecular dynamics of initial structures embedded into the water solution environment were performed. Multiple-walkers metadynamics was also used to enhance the sampling of structures encompassing both functional and non-functional organizations of iron-sulfur clusters. The resulting structural model provided by deep learning is consistent with functional [FeFe] hydrogenase characterized by peculiar interactions between cofactors and the protein matrix.

Abstract 4921: A user-friendly R Shiny web app for predicting cancer genetic dependencies using deep learning

Abstract Understanding the genetic vulnerabilities (i.e., genetic dependencies) of cancer cells is crucial for developing novel anti-cancer treatments. The Cancer Dependency Map (DepMap) projects, conducted by the Broad Institute and Wellcome Sanger Institute, have performed comprehensive genome-wide CRISPR-Cas9 knockout screens across diverse cancer cell lines. These data resources have enabled computational biologists to glean insights into cancer genetic dependencies using sophisticated computational models. As different genomic mechanisms govern tumorigenesis and genetic vulnerability, the relationship between cancer genomics and genetic dependencies is nonlinear, making the prediction of genetic dependencies challenging. Addressing this challenge, we previously developed a deep learning model, namely DeepDEP, to predict cancer genetic dependencies using multi-omic profiles (Chiu et al. DOI:10.1126/sciadv.abh1275). DeepDEP has a transfer learning scheme that integrates genomic representations captured from unlabeled tumor samples of the Cancer Genome Atlas (TCGA) and genomic features that are predictive of genetic dependencies in cell lines. Such a design enables the application of DeepDEP to both cell lines and tumors. The model demonstrated superior performance compared to various conventional machine learning methods. Prediction results were validated using independent cell-line datasets and patient clinical data. Here, we developed an intuitive web application for DeepDEP which allows users to easily utilize the model on their in-house data. Users can simply indicate whether the query sample is derived from a cancer cell line or a tumor, upload any combination of mutation, gene expression, DNA methylation, and copy number alteration datasets, and view the predicted dependencies of 1,298 cancer-relevant genes. Along with the genetic dependency predictions, users are provided with model performance metrics, detailed gene annotations, and links to useful resources such as Ensembl and COSMIC. Our web application features a variety of interactive visualizations to guide users to interpret the results at the levels of individual genes and pathways. The tool has two supplemental modules to enhance its functionality. One allows users to search for similar cell lines to the query sample within the DepMap project or for similar tumors within the TCGA project. The other supplemental module provides data exploration tools for pre-calculated predictions across all TCGA tumors. In summary, this user-friendly web application provides researchers with access to deep learning predictions of cancer genetic dependencies with various interactive visualization and analysis tools. It is anticipated that this application will facilitate the study of cancer genetic dependencies and enhance the efficiency of anti-cancer drug discovery. Citation Format: Michael J. Kasper, Li-Ju Wang, Michael Ning, Yufei Huang, Yu-Chiao Chiu. A user-friendly R Shiny web app for predicting cancer genetic dependencies using deep learning [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4921.

Deep learning for monthly rainfall–runoff modelling: a large-sample comparison with conceptual models across Australia

Abstract. A deep learning model designed for time series predictions, the long short-term memory (LSTM) architecture, is regularly producing reliable results in local and regional rainfall–runoff applications around the world. Recent large-sample hydrology studies in North America and Europe have shown the LSTM model to successfully match conceptual model performance at a daily time step over hundreds of catchments. Here we investigate how these models perform in producing monthly runoff predictions in the relatively dry and variable conditions of the Australian continent. The monthly time step matches historic data availability and is also important for future water resources planning; however, it provides significantly smaller training datasets than daily time series. In this study, a continental-scale comparison of monthly deep learning (LSTM) predictions to conceptual rainfall–runoff (WAPABA model) predictions is performed on almost 500 catchments across Australia with performance results aggregated over a variety of catchment sizes, flow conditions, and hydrological record lengths. The study period covers a wet phase followed by a prolonged drought, introducing challenges for making predictions outside of known conditions – challenges that will intensify as climate change progresses. The results show that LSTM models matched or exceeded WAPABA prediction performance for more than two-thirds of the study catchments, the largest performance gains of LSTM versus WAPABA occurred in large catchments, the LSTMs struggled less to generalise than the WAPABA models (e.g. making predictions under new conditions), and catchments with few training observations due to the monthly time step did not demonstrate a clear benefit with either WAPABA or LSTM.

Making Early and Accurate Deep Learning Predictions to Help Disadvantaged Individuals in Medical Crowdfunding

Medical crowdfunding is a popular channel for people seeking financial assistance to cover their medical expenses, allowing them to collect donations from a large number of donors. However, a mismatch between the supply and demand of donations creates large heterogeneity in the fundraising outcomes across medical crowdfunding campaigns, and such uncertainty can impede the timely planning of treatment for patients. Providing early and accurate forecasts for medical crowdfunding performance can better inform fundraisers and assist them in optimizing timely interventions to improve fundraising outcomes. In this study, we propose a new approach that effectively combines time-varying features and time-invariant features in a deep learning model, to provide dynamic predictions of fundraising outcomes. When compared with a comprehensive set of baseline models, our model consistently demonstrates higher predictive accuracy while requiring a shorter observation window of data, thus achieving both accurate and early prediction objectives. We further conduct a temporal clustering analysis to analyze the heterogeneous patterns in how the time-varying features relate to fundraising outcomes. In addition, we perform simulation analyses to demonstrate that interventions from fundraisers can significantly improve the fundraising performance of disadvantaged cases that are predicted to receive the lowest donation amounts, particularly when the interventions are implemented early. These findings show that our deep learning prediction model and the actionable insights can provide timely feedback to fundraisers and promote equal access to resources for all. Our proposed approach is applicable to various contexts, enabling effective processing of diverse sources of data and facilitating early interventions.