Comprehensive Sensitivity Analysis Research Articles

Causal role of myeloid cells in Parkinson's disease: Mendelian randomization study.

Previous studies have observed elevated myeloid cells in the peripheral blood of patients with Parkinson's disease (PD), but the causal relationship between them remains to be elucidated. We investigated whether there is a causal relationship between different subtypes of peripheral blood myeloid cells and PD using Mendelian randomization (MR) combined with bioinformatics analysis. Exploring the etiology of PD from the perspective of genetics can remove confounding factors and provide a more reliable theoretical basis for elucidating the pathogenesis of PD. Comprehensive two-sample MR analysis and sensitivity analyses were conducted to explore the causal associations between 64 myeloid cell signatures and PD risk. The Venn diagram and protein-protein interaction network analysis of instrumental variables (IV) corresponding genes were used to further investigate the potential mechanism of myeloid cells influencing the pathogenesis of PD. We investigated the impact of four immunophenotypes on the risk of PD, including Im MDSC% CD33dim HLA DR- CD66b- (relative count), CD33dim HLA DR+ CD11b+% CD33dim HLA DR+ (relative count), and CD11b on Mo MDSC (MFI) and CD11b on CD33br HLA DR+ CD14dim (MFI), while an immunophenotype's protective effect on PD was observed CD45 on Im MDSC (MFI). The results of bioinformatics analysis showed that CD33, NTRK2, PLD2, GRIK2 and RELN had protein interactions with the risk genes of PD. Our study has demonstrated a close genetic correlation between different subtypes of myeloid cells and PD, providing guidance for early identification and immunotherapeutic development in patients with PD.

Sustainable strategies based on the social responsibility of the beverage industry companies for the circular supply chain

Circular supply chain management (CSCM) offers a fresh approach to enhancing supply chain sustainability and minimizing waste. A significant amount of waste is generated daily in the beverage industry, primarily due to the extensive production of beverages. The increasing waste amount causes the destruction of the environment and many problems for human life. The social responsibility of the beverage industry dictates that social and environmental performances are combined with the company’s economic performance to benefit society and the environment. The purpose of this research is to evaluate the effective factors and provide sustainable solutions for waste management in the CSCM of beverage industry companies based on their social responsibility. So in this regard, six sustainable strategies are suggested and evaluated based on the circular design, biodegradable packaging, product and manufacturer’s responsibility, critical success factors, and corporate social responsibility criteria, and theirs 24 sub-criteria for improving CSCM in the beverage industry. A novel group decision-making approach is proposed by developing the base-criterion method (BCM) and multi-attribute border approximation area comparison (MABAC) under fuzzy Z-extended numbers in order to evaluate the criteria weights and the rank of the strategies. A comprehensive managerial sensitivity analysis was performed better to understand the impact of different criteria for each strategy. The results show that cooperation with charities to return the waste and spend the added value created to help the needy fulfill the company’s social responsibility is one of the most important strategies specified by decision-makers and experts for improving the CSCM of beverage industry.

Source term inversion coupling Kernel Principal Component Analysis, Whale Optimization Algorithm, and Backpropagation Neural Networks (KPCA-WOA-BPNN) for complex dispersion scenarios

Accurate and rapid source term estimation is critical for consequence assessment and emergency decision-making in nuclear accidents. Neural network methods provide a promising approach to achieving this goal, but they are mainly demonstrated in ideal scenarios with flat terrain. In this study, a source term inversion model combining the Kernel Principal Component Analysis, the Whale Optimization Algorithm, and the Backpropagation Neural Networks (KPCA-WOA-BPNN) was proposed for source term inversion in a more realistic scenario with complex terrain. The method was validated against two simulated release scenarios based on the meteorology and topography of a real Chinese nuclear power plant. A comprehensive sensitivity analysis of the position distribution, quantity, and quality of measurement sites and the four input parameters was conducted. The results revealed that the relative error of the KPCA-WOA-BPNN was below 2% for both scenarios. The sites within 1.2 km of the release point are critical for the performance. Even with only 6 sites, the error of the method can be as low as 3.82%. The method is insensitive to the disturbances in wind speed, mix layer height, and release height, but less robust to those in concentration. The method also works with other optimization algorithms and presents low error levels, indicating good extendibility.

Northern Lights: Prospecting Efficiency in Europe’s Renewable Energy Sector

Northern European nations are at the forefront of renewable energy adoption but face challenges in optimizing energy conversion efficiency. There is a lack of detailed understanding of how behavioral factors affect the efficiency of renewable energy conversion in these countries. This study aims to evaluate and compare the renewable energy conversion efficiency of Northern European countries, intending to inform strategic policy making and identify best practices for technology deployment in the renewable energy sector. Employing a Data Envelopment Analysis (DEA) model, the study integrates behavioral economic parameters—specifically, the aversion loss and gain significance coefficients—to assess the efficiency of renewable energy conversion, accounting for psychological factors in decision making. A comprehensive sensitivity analysis was conducted, varying the gain significance coefficient while maintaining the aversion loss coefficient at constant levels. This experiment was designed to observe the impact of behavioral parameters on the efficiency ranking of each country. The analysis revealed that Latvia consistently ranked highest in efficiency, irrespective of the gain significance valuation, whereas Iceland consistently ranked lowest. Other countries demonstrated varying efficiency rankings with changes in gain significance, indicating different behavioral economic influences on their renewable energy sectors. Theoretically, the study enhances the DEA framework by integrating behavioral economics, offering a more holistic view of efficiency in renewable energy. Practically, it provides a benchmarking perspective that can guide policy and investment in renewable energy, with sensitivity analysis underscoring the importance of considering behavioral factors. The research offers a practical tool for policymakers and energy stakeholders to align renewable energy strategies with behavioral incentives, aiming to improve the adoption and effectiveness of these initiatives.

Causal relationship between immune cells and prostate cancer: a Mendelian randomization study.

Despite the abundance of research indicating the participation of immune cells in prostate cancer development, establishing a definitive cause-and-effect relationship has proven to be a difficult undertaking. This study employs Mendelian randomization (MR), leveraging genetic variables related to immune cells from publicly available genome-wide association studies (GWAS), to investigate this association. The primary analytical method used in this study is inverse variance weighting (IVW) analysis. Comprehensive sensitivity analyses were conducted to assess the heterogeneity and horizontal pleiotropy of the results. The study identifies four immune cell traits as causally contributing to prostate cancer risk, including CD127- CD8+ T cell %CD8+ T cell (OR = 1.0042, 95%CI:1.0011-1.0073, p = 0.0077), CD45RA on CD39+ resting CD4 regulatory T cell (OR = 1.0029, 95%CI:1.0008-1.0050, p = 0.0065), CD62L- Dendritic Cell Absolute Count (OR = 1.0016; 95%CI:1.0005-1.0026; p = 0.0039), CX3CR1 on CD14+ CD16- monocyte (OR = 1.0024, 95%CI:1.0007-1.0040, p = 0.0060). Additionally, two immune cell traits are identified as causally protective factors: CD4 on monocyte (OR = 0.9975, 95%CI:0.9958-0.9992, p = 0.0047), FSC-A on plasmacytoid Dendritic Cell (OR = 0.9983, 95%CI:0.9970-0.9995, p = 0.0070). Sensitivity analyses indicated no horizontal pleiotropy. Our MR study provide evidence for a causal relationship between immune cells and prostate cancer, holding implications for clinical diagnosis and treatment.

Storm surge risk assessment and sensitivity analysis based on multiple criteria decision-making methods: a case study of Huizhou City

Coastal areas face escalating storm surge disasters due to rising sea levels and urban growth, posing greater risks to lives and property. Comprehensive storm surge risk assessment and sensitivity analysis in coastal areas are crucial for effective disaster prevention and mitigation. This research focuses on Huizhou, China, conducting a comparative study of storm surge risk assessment and sensitivity analysis based on an integrated approach, which includes the storm surge inundation numerical model (FVCOM-SWAN), Geographic Information System (GIS) and Remote Sensing (RS) techniques, and Multiple Criteria Decision-Making (MCDM) methods. Ten flood-related risk indicators are selected from the hazard, exposure and vulnerability, the weights of which are evaluated through a comprehensive comparison among Analytic Hierarchy Process (AHP), Fuzzy Analytic Hierarchy Process (FAHP), Entropy Weight (EW), AHP-EW, and FAHP-EW methods. High-precision risk level maps are generated subsequently utilizing GIS and RS techniques. Sensitivities of the indicators are analyzed using One-At-A-Time (OAT) and Fourier Amplitude Sensitivity Test (FAST) methods. The proposed storm surge risk assessment framework, the MCDM comparative study and the sensitivity analysis can offer insights for better understanding and management of storm surge risks, and contribute to the standardization and application of storm surge risk assessment.

Electrochemical aging model of lithium-ion battery with impedance output and its parameter sensitivity analysis and identification

The electrochemical model of the battery has gained widespread recognition due to its clear physical interpretability. However, achieving an accurate electrochemical model heavily relies on precise parameter settings. To address this challenge, various model-based parameter optimization methods have been developed. Currently, most research relies on model reduction techniques and heuristic random search methods for parameter identification. These approaches often lead to time-consuming processes and a loss of physical interpretability in the electrochemical model. To overcome these limitations, we propose a novel technical framework that directly parametrizes the electrochemical model. First, we establish a battery electrochemical model that accounts for lithium plating and SEI film thickening, serving as the main focus of this study. Secondly, in order to expedite parameter identification without compromising the model's physical interpretability, we innovate by performing comprehensive sensitivity analysis using voltage and impedance characteristics. This allows us to identify the most influential parameters, which become the key for subsequent research. Finally, to ensure reasonable initialization of finite element computations and to avoid convergence issues in the identification algorithm, we combine machine learning and optimization algorithms into a two-step parameter identification method. By utilizing the proposed method to identify the model parameters, the simulations exhibit high accuracy.

FishSegSSL: A Semi-Supervised Semantic Segmentation Framework for Fish-Eye Images.

The application of large field-of-view (FoV) cameras equipped with fish-eye lenses brings notable advantages to various real-world computer vision applications, including autonomous driving. While deep learning has proven successful in conventional computer vision applications using regular perspective images, its potential in fish-eye camera contexts remains largely unexplored due to limited datasets for fully supervised learning. Semi-supervised learning comes as a potential solution to manage this challenge. In this study, we explore and benchmark two popular semi-supervised methods from the perspective image domain for fish-eye image segmentation. We further introduce FishSegSSL, a novel fish-eye image segmentation framework featuring three semi-supervised components: pseudo-label filtering, dynamic confidence thresholding, and robust strong augmentation. Evaluation on the WoodScape dataset, collected from vehicle-mounted fish-eye cameras, demonstrates that our proposed method enhances the model's performance by up to 10.49% over fully supervised methods using the same amount of labeled data. Our method also improves the existing image segmentation methods by 2.34%. To the best of our knowledge, this is the first work on semi-supervised semantic segmentation on fish-eye images. Additionally, we conduct a comprehensive ablation study and sensitivity analysis to showcase the efficacy of each proposed method in this research.

Dietary fiber intake and its association with diabetic kidney disease in American adults with diabetes: A cross-sectional study.

Dietary fiber (DF) intake may have a protective effect against type 2 diabetes (T2D); however, its relationship with diabetic kidney disease (DKD) remains unclear. To investigate the potential association between DF intake and the prevalence of DKD in individuals diagnosed with T2D. This cross-sectional study used data from the National Health and Nutrition Examination Survey collected between 2005 and 2018. DF intake was assessed through 24-h dietary recall interviews, and DKD diagnosis in individuals with T2D was based on predefined criteria, including albuminuria, impaired glomerular filtration rate, or a combination of both. Logistic regression analysis was used to assess the association between DF intake and DKD, and comprehensive subgroup and sensitivity analyses were performed. Among the 6032 participants, 38.4% had DKD. With lower DF intake-T1 (≤ 6.4 g/1000 kcal/day)-as a reference, the adjusted odds ratio for DF and DKD for levels T2 (6.5-10.0 g/1000 kcal/day) and T3 (≥ 10.1 g/1000 kcal/day) were 0.97 (95%CI: 0.84-1.12, P = 0.674) and 0.79 (95%CI: 0.68-0.92, P = 0.002), respectively. The subgroup analysis yielded consistent results across various demographic and health-related subgroups, with no statistically significant interactions (all P > 0.05). In United States adults with T2D, increased DF intake may be related to reduced DKD incidence. Further research is required to confirm these findings.

Condition-based maintenance policy for systems under dynamic environment

This paper addresses the challenge of optimizing maintenance for systems with varying degradation characteristics in dynamic environment. The focus is on repairable systems, where degradation is characterized by a Wiener process that incorporates the dynamic influence of the environment. We investigate a condition-based maintenance model aiming at minimizing the expected cost over a finite and infinite horizon. Firstly, the model considers the evolution of environment as a Markov process, which influences the drift parameter in the Wiener process. The choice between a corrective and preventive replacement rests on the periodic inspections on the system and environment. Subsequently, the maintenance model is formulated in the framework of Markov decision process. Meanwhile, the backward dynamic programming algorithm and value iteration algorithm are employed to gain the optimal maintenance policies over the finite and the infinite planning horizons, respectively. Finally, the proposed model is further explained with concrete examples and a comprehensive sensitivity analysis. Numerical results clearly demonstrate notable variations in the optimal maintenance strategies in different periods and underscore the importance of taking proactive maintenance measures in harsh environmental conditions, even if the degradation level of the system is relatively low.

Spatiotemporal distribution of carbon monoxide dynamics and probabilistic health risk assessment: A comprehensive sensitivity analysis across India's state capitals

Carbon monoxide (CO) is a harmful trace gas produced from the incomplete combustion of fossil fuels. Carbon monoxide (CO) poses a substantial threat to human health. The objectives of this study were to investigate carbon monoxide (CO) annual and seasonal variations with meteorological factors, study the marginal effects of weather factors on CO, analyze the impact of independent variables (meteorological factors) importance on CO distribution, time series trend analysis (Mann-Kendall and Sen's slope, additive decomposition model, and linear trend model), study the spatial distribution of CO in the capital cities and non-carcinogenic health risk in India from 2019 to 2022. The study showed that CO concentrations varied, notably across different years (from 2019 to 2022), months, and seasons. The maximum annual CO (0.91 ± 0.17 mg/m3) concentration was recorded in 2019, while the lowest was (0.67 ± 0.12 mg/m3) in 2022. Seasonally, the winter season had the highest load of CO (ranging from 0.17 to 2.17 mg/m3) in the ambient atmosphere compared to the monsoon (0.08 to 1.38 mg/m3) and summer (0.10 to 1.50 mg/m3) seasons. The analysis of the importance of the independent variables (meteorological factors) on CO indicated that temperature had the most substantial impact on CO variation, followed by wind direction, relative humidity, and wind speed. The trend analysis using different models showed a declining trend in CO concentrations over the study period. The non-carcinogenic health risk HQ values were less than 1, suggesting no notable risk. The Monte Carlo simulation revealed that the 5th and 95th percentile range of HQ risk was 0.02 to 0.07 and the sensitivity analysis of non-carcinogenic risk at the state capital level was conducted, and revealed that Delhi was at a higher risk than Chandigarh, Patna, Gandhinagar, Bengaluru, Mumbai, and others, respectively. The results of this study on CO emissions will be beneficial in creating focused plans and policies to reduce CO pollution in seasonal influences and its effects on ambient air quality and public health in the capital cities of Indian states.

An integrated multi-criteria approach to formulate and assess healthcare referral system strategies in developing countries

This study aims to identify challenges in implementing a quality healthcare referral system in developing countries and explore the strategies to overcome these challenges. Data for this study were collected through consultations with experts in the field. We introduce a novel hybrid method called Criteria Importance Assessment (CIMAS) and Alternative Ranking Order Method Accounting for Two-Step Normalization (AROMAN). CIMAS determines the relative importance of criteria, and AROMAN is employed to rank the strategies. The primary challenges identified include inadequate infrastructure facilities and deficient health information systems. The most appropriate strategy involves focusing on improving infrastructure facilities. We also carry out comprehensive sensitivity and comparative analyses to validate the applicability of the proposed model. This study identifies and elucidates the challenges of establishing a high-quality healthcare referral system in developing countries and substantially contributes to the existing body of knowledge by effectively delineating and prioritizing the strategies to tackle these challenges.

Machine learning for the prediction of local asteroid damages

Risk assessment studies of local asteroid hazards traditionally simulate the physics of meteors with engineering models tailored to analyze tens-of-millions of scenarios. However, these simplified approaches still need to solve time-dependent ODEs to model the entry process and the resulting ground damage. With a computational cost of O(0.01 CPU.s) per scenario, simulating these large numbers of potential entry conditions can take several days on local computers. To improve computational efficiency, we propose in this paper an orthogonal approach based on machine learning models to predict the size of damaged areas given a list of entry parameters. We train 5 machine learning methods and compare the predictions to the outputs of NASA’s state-of-the-art Probabilistic Asteroid Impact Risk (PAIR) model. PAIR’s remarkable capability to handle a broad spectrum of entry conditions and a high volume of scenarios establishes it as our reference model, with PAIR estimates assumed ground truth and used to train, test, and validate the machine learning models. The ML models, initially trained with PAIR’s primitive input variables, are then further enhanced with engineered features to improve the performance of interpretable models. We find that complex models like neural networks are well-suited to estimate blast hazards, while simpler, interpretable linear models can accurately assess thermal damage. For both types of hazards, the radii of damaged areas can be predicted with less than 10% average relative errors for radii greater than 100 km. The CPU time is decreased by a factor O(103) compared to the PAIR model, enabling the simulation of millions of scenarios within minutes on a local computer. Beyond serving as surrogate models, we incorporate the machine learning models into a comprehensive Shapley sensitivity analysis, yielding a ranking of entry parameters based on their contributions to ground damages. These results offer valuable insights for prioritizing observation missions.

Tax revenue lost due to illicit cigarettes in South Africa: 2002−2022

ObjectiveTo estimate lost excise and value-added tax (VAT) revenue as a result of illicit cigarette trade from 2002 to 2022.DesignUsing gap analysis, we estimated the number of illicit cigarettes by...

Railway Signaling System Interference by Unbalanced Traction Harmonic Current: Mechanism, Modeling, and Solutions Based on Engineering Practice

The rapid development of railway electrification has led to signaling relay malfunctions and locomotive urgent braking accidents caused by unbalanced traction harmonic current (UTHC), which seriously threaten transit safety. This paper presents a comprehensive study with a mechanism-modeling-solution approach that addresses the practical issue of UTHC interference in signaling systems and provides theoretical support for further research, relying on an actual case. Based on the mechanism analysis of UTHC interfering with the on-board and ground signaling terminals, UTHC and its induced harmonic electromotive force on both terminals are accurately modeled. The proposal and evaluation of solutions are supported by mathematical analysis of the relationship between traction-side current and signaling-side induced electromotive force. A low active loss passive filter with high filtering effect was designed and connected to the traction network for the actual case, effectively addressing the high-frequency harmonic resonance risk in traction power supply systems and the harmonic interference issue in signaling systems. Field tests validate the feasibility and effectiveness of this solution. Accounting for cost constraints, guidelines for reducing current unbalance between two rails or avoiding choke transformer core saturation are developed based on a comprehensive sensitivity analysis of traction current unbalance coefficient and signaling equipment parameters.

Genetic effects of inflammatory cytokines on coronary artery disease and myocardial infarction and the mediating roles of lipid traits.

Chronic inflammation has been connected by epidemiological evidence to coronary artery disease (CAD) along with myocardial infarction (MI). Nevertheless, it is still unclear whether reverse causality or confounders account for these connections. Our objectives are to examine the causality between inflammatory cytokines and CAD/MI as well as the potential mediating influence of lipid characteristics. We acquired instrumental variables through genome-wide association studies meta-analyses of 41 inflammatory cytokines (8293 individuals). Genetic associations with CAD (122 733 cases and 424 528 controls), MI (~61 505 cases and 577 716 controls) and five candidate lipid mediators were obtained from the corresponding genome-wide association studies. A two-step, two-sample Mendelian randomization analysis was applied, followed with comprehensive sensitivity analyses. Genetically determined growth regulated oncogene-α was causally linked to a decreased incidence of CAD [odds ratio (OR), 0.97; 95% confidence interval (CI), 0.95-0.99; P = .007] and MI (OR, 0.95; 95% CI, 0.92-0.98; P = .002). There is suggestive evidence indicating a causal impact of macrophage inflammatory protein-1β upon CAD (OR, 1.04; 95% CI, 1.01-1.07; P = .010) and MI (OR, 1.07; 95% CI, 1.02-1.11; P = .002). Furthermore, we discovered suggestive causal connections between tumor necrosis factor-related apoptosis-inducing ligand and CAD (OR, 0.97; 95% CI, 0.95-1.00; P = .020). Two-step Mendelian randomization analysis revealed that triglycerides partially mediate the effect of growth regulated oncogene-α on CAD (proportion-mediated: 13.28%) and MI (8.05%). We provided novel genetic evidence supporting the causality of inflammatory cytokines on CAD/MI and elucidate the mediating effect of triglycerides in the causal pathways linking inflammatory cytokines and CAD/MI.

Comparative analysis of soft-error sensitivity in LU decomposition algorithms on diverse GPUs

Graphics processing units (GPUs) have become integral to embedded systems and supercomputing centres due to their large memory, cutting-edge technology and high performance per watt. However, their susceptibility to transient errors requires a comprehensive analysis of error sensitivity, as well as the development of error mitigation techniques and fault-tolerant algorithms. This study focuses on evaluating the soft-error sensitivity of two distinct versions of LU decomposition algorithms implemented on two very different GPUs—a low-power SoC embedded GPU and a high-performance massively parallel GPU. Through extensive fault injection campaigns on both GPUs, we examine the vulnerability of the algorithms, identify error causes, and determine critical code components requiring enhanced protection. The experiments reveal that most single bit flip fault injections in the instruction results lead to erroneous outcomes or unrecoverable errors. Notably, efficient GPU resource utilisation can increase the number of masked errors, thereby enhancing error resilience. Additionally, while different parts of the code exhibit similar error occurrence types and rates, the propagation of errors to elements within the result matrix differs significantly.

Causal relationship between psychological factors and hepatocellular carcinoma as revealed by Mendelian randomization

PurposeThe impact of psychological factors on the incidence of hepatocellular carcinoma (HCC) in humans remains unclear. Mendelian randomization (MR) study is a novel approach aimed at unbiased detection of causal effects. Therefore, we conducted a two-sample MR to determine if there is a causal relationship between psychological distress (PD), participation in leisure/social activities of religious groups (LARG), and HCC.MethodsThe genetic summary data of exposures and outcome were retrieved from genome-wide association studies (GWAS). We used PD and LARG as exposures and HCC as outcome. Five MR methods were used to investigate the causal relationship between PD, LARG, and HCC. The result of inverse variance weighted (IVW) method was deemed as principal result. Besides, we performed a comprehensive sensitivity analysis to verify the robustness of the results.ResultsThe IVW results showed that PD [odds ratio (OR) 1.006, 95% confidence interval (CI) 1.000–1.011, P = 0.033] and LARG (OR 0.994, 95% CI 0.988–1.000, P = 0.035) were causally associated with the incidence of HCC. Sensitivity analysis did not identify any bias in the results.ConclusionPD turned out to be a mild risk factor for HCC. In contrast, LARG is a protective factor for HCC. Therefore, it is highly recommended that people with PD are seeking positive leisure activities such as participation in formal religious social activities, which may help them reduce the risk of HCC.Graphical abstract

Optimizing the Journey: Dynamic Charging Strategies for Battery Electric Trucks in Long-Haul Transport

Battery electric trucks (BETs) represent a well-suited option for decarbonizing road freight transport to achieve climate targets in the European Union. However, lower ranges than the daily distance of up to 700 km make charging stops mandatory. This paper presents an online algorithm for optimal dynamic charging strategies for long-haul BET based on a dynamic programming approach. In several case studies, we investigate the advantages optimal strategies can bring compared to driver decisions. We further show which charging infrastructure characteristics in terms of charging power, density, and charging station availability should be achieved for BETs in long-haul applications to keep the additional time required for charging stops low. In doing so, we consider the dynamic handling of occupied charging stations for the first time in the context of BET. Our findings show that, compared to driver decisions, optimal charging strategies can reduce the time loss by half compared to diesel trucks. To keep the time loss compared to a diesel truck below 30 min a day, a BET with a 500 kWh battery would need a charging point every 50 km on average, a distributed charging power between 700 and 1500 kW, and an average charger availability above 75%. The presented method and the case studies’ results’ plausibility are interpreted within a comprehensive sensitivity analysis and subsequently discussed in detail. Finally, we transformed our findings into concrete recommendations for action for the efficient rollout of BETs in long-haul applications.

Comprehensive Sensitivity Analysis Framework for Transfer Learning Performance Assessment for Time Series Forecasting: Basic Concepts and Selected Case Studies

Recently, transfer learning has gained popularity in the machine learning community. Transfer Learning (TL) has emerged as a promising paradigm that leverages knowledge learned from one or more related domains to improve prediction accuracy in a target domain with limited data. However, for time series forecasting (TSF) applications, transfer learning is relatively new. This paper addresses the need for empirical studies as identified in recent reviews advocating the need for practical guidelines for Transfer Learning approaches and method designs for time series forecasting. The main contribution of this paper is the suggestion of a comprehensive framework for Transfer Learning Sensitivity Analysis (SA) for time series forecasting. We achieve this by identifying various parameters seen from various angles of transfer learning applied to time series, aiming to uncover factors and insights that influence the performance of transfer learning in time series forecasting. Undoubtedly, symmetry appears to be a core aspect in the consideration of these factors and insights. A further contribution is the introduction of four TL performance metrics encompassed in our framework. These TL performance metrics provide insight into the extent of the transferability between the source and the target domains. Analyzing whether the benefits of transferred knowledge are equally or unequally accessible and applicable across different domains or tasks speaks to the requirement of symmetry or asymmetry in transfer learning. Moreover, these TL performance metrics inform on the possibility of the occurrence of negative transfers and also provide insight into the possible vulnerability of the network to catastrophic forgetting. Finally, we discuss a sensitivity analysis of an Ensemble TL technique use case (with Multilayer Perceptron models) as a proof of concept to validate the suggested framework. While the results from the experiments offer empirical insights into various parameters that impact the transfer learning gain, they also raise the question of network dimensioning requirements when designing, specifically, a neural network for transfer learning.