Uplift Model Research Articles

Conversion Uplift in E-Commerce: A Systematic Benchmark of Modeling Strategies

Uplift modeling combines machine learning and experimental strategies to estimate the differential effect of a treatment on individuals’ behavior. The paper considers uplift models in the scope of marketing campaign targeting. Literature on uplift modeling strategies is fragmented across academic disciplines and lacks an overarching empirical comparison. Using data from online retailers, we fill this gap and contribute to literature through consolidating prior work on uplift modeling and systematically comparing the predictive performance and utility of available uplift modeling strategies. Our empirical study includes three experiments in which we examine the interaction between an uplift modeling strategy and the underlying machine learning algorithm to implement the strategy, quantify model performance in terms of business value and demonstrate the advantages of uplift models over response models, which are widely used in marketing. The results facilitate making specific recommendations how to deploy uplift models in e-commerce applications.

Numerical Modeling of Weathering, Erosion, Sedimentation, and Uplift in a Triple Junction Divergent Margin

AbstractThe majority of numerical models of landscape evolution in divergent margins are focused on the simulation of margins with simplified lithological control on landscape erosion. However, this approach is insufficient to study the evolution of margins where chemical weathering is an important element increasing rock resistance to physical erosion. One example of this margin is the Borborema Province, northeastern Brazil, where postrift marine sediments are now preserved at elevations ∼700–800 m in up to 1‐km‐high plateaus capped by duricrust layers. The landscape evolution of these uplifted sedimentary basins still eludes explanation. Here we use numerical models that couple weathering, erosion, sedimentation, sea level changes, flexural isostasy, and thermal effects due to lithospheric stretching to simulate the tectonosedimentary evolution of the Borborema Province since the onset of continental stretching during the Lower Cretaceous. These numerical experiments reveal that nearly 70% of the postrift regional uplift observed in the Borborema Province can be explained by differential denudation of the continent and flexural rebound of the lithosphere. The remaining ∼250 m of uplift can be explained by thermal uplift induced by partial erosion of the base of the continental lithosphere under the Borborema Province due to edge‐driven convection, in accordance with the anomalously thin continental lithosphere observed under the Borborema Province. Additionally, the numerical results can explain the regional pattern of fission track ages by the combined effect of differential denudation and flexural rebound in this geometrically complex margin.

4D models of de Sitter uplift

It was shown in arXiv:1808.09428 that the modified 4d version of the KKLT model proposed in arXiv:1707.08678 is inconsistent for large values of the parameter $c$ advocated in arXiv:1707.08678, since there is a point in the moduli space where $|D_SW|^2$ vanishes. The authors responded with yet another modification of the 4d KKLT model arXiv:1809.06618. However, for large $c$, this model suffers from an even worse problem: not only is there a point in the moduli space where $|D_SW|^2$ vanishes, there is also a region in the moduli space where $|D_SW|^2$ is negative. Meanwhile for small $c$ these models have dS vacua. We construct improved models, which are fully consistent for all values of parameters, just as the original version of the KKLT model using a nilpotent superfield. These models have a family of dS vacua for a broad range of parameter values. Thus, the results of the analysis of all presently available consistent generalizations of the 4d KKLT model, in the domain of their validity, confirm the existence of dS vacua in the KKLT scenario.

Response Transformation and Profit Decomposition for Revenue Uplift Modeling

Uplift models support decision-making in marketing campaign planning. Estimating the causal effect of a marketing treatment, an uplift model facilitates targeting communication to responsive customers and efficient allocation of marketing budgets. Research into uplift models focuses on conversion models to maximize incremental sales. The paper introduces uplift modeling strategies for maximizing incremental revenues. If customers differ in their spending behavior, revenue maximization is a more plausible business objective compared to maximizing conversions. The proposed methodology entails a transformation of the prediction target, customer-level revenues, that facilitates implementing a causal uplift model using standard machine learning algorithms. The distribution of campaign revenues is typically zero-inflated because of many non-buyers. Remedies to this modeling challenge are incorporated in the proposed revenue uplift strategies in the form of two-stage models. Empirical experiments using real-world e-commerce data confirm the merits of the proposed revenue uplift strategy over relevant alternatives including uplift models for conversion and recently developed causal machine learning algorithms. To quantify the degree to which improved targeting decisions raise return on marketing, the paper develops a decomposition of campaign profit. Applying the decomposition to a digital coupon targeting campaign, the paper provides evidence that revenue uplift modeling, as well as causal machine learning, can improve campaign profit substantially.

The Importance of the Inelastic and Elastic Structures of the Crust in Constraining Glacial Density, Mass Change, and Isostatic Adjustment From Geodetic Observations in Southeast Alaska

AbstractElastic deformation of the solid Earth in response to ice mass loss offers a promising constraint on the density of glacial material lost. Further, the elastic response to modern deglaciation is important to constrain for studies of glacial isostatic adjustment to determine the mantle's structure and rheology. Models of this elastic uplift are commonly based on the 1‐D, seismically derived global average Preliminary Reference Earth Model and typically neglect uncertainties that can arise from regional differences in elastic structure from that of the global average, lateral heterogeneities within the region, and inelastic behavior of the crust. We quantify these uncertainties using an ensemble of 1‐D local elastic structure models and empirical relations for the effects of inelasticity in the upper ∼10 km of the crust. In Southeast Alaska, modeling elastic uplift rates with local elastic structures results in up to a 20–40% difference from those modeled with the Preliminary Reference Earth Model. Although these differences are limited to regions near to ice‐covered areas, they are comparable to the differences in uplift rates expected from the loss of firn versus loss of ice. Far from ice‐covered areas, where most of the region's GPS observations were made, these differences become insignificant and do not affect previous glacial isostatic adjustment studies in the region. The methods presented here are based on the globally available LITHO1.0 seismic model and open source software, and the approach of using an ensemble of 1‐D elastic structures can be easily adapted to other regions around the world.

Modelling and visualising distributed crustal deformation of Australia and Zealandia using GPlates 2.0

The recently released GPlates 2.0 software (www.gplates.org) provides a framework for building plate motion models including distributed extension and compression, driven by the motions of the surrounding rigid plate interiors, with constraints assimilated from well and seismic data. Here we present a regional deforming plate model for Australia and Zealandia. It captures the progressive extension of all Australian continental margins, starting with the Jurassic extension of the northwest shelf, and the Cretaceous extension of the southern and eastern Australian margins. The model also includes the extension of the Lord Howe Rise and southern Zealandia in the mid- to Late Cretaceous and the subsequent complex compressional deformation of New Zealand since the early Miocene. The model enables the time-dependent computation of crustal stretching factors for passive margins, as well as compression factors for orogenic regions. The thinning/thickening can be tracked by points distributed within a deforming mesh, either by starting with an assumed initial stretching factor or by applying an iterative approach in estimating initial crustal thicknesses to account for present-day constraints. The latter is suitable for basins, while the former is more applicable for orogens, where the present-day crustal thickness is not a reliable indicator of total crustal thickening due to the effects of erosion. Modelled crustal stretching factors can also be ground-truthed with independently derived stretching factors from strain rate inversion of stratigraphic sequences from well data or forward models of tectonic subsidence. The model can be combined with estimates of mantle-driven dynamic topography through time to generate basement subsidence or uplift models that include isostatic and dynamic components, serving as boundary conditions for basin and source-to-sink sediment transport models to provide improved constraints for resource exploration.

Quaternary uplift of the northern margin of the Central Anatolian Plateau: New OSL dates of fluvial and delta-terrace deposits of the Kızılırmak River, Black Sea coast, Turkey

We analysed the interplay between coastal uplift, sea level change in the Black Sea, and incision of the Kızılırmak River in northern Turkey. These processes have created multiple co-genetic fluvial and marine terrace sequences that serve as excellent strain markers to assess the ongoing evolution of the Pontide orogenic wedge and the growth of the northern margin of the Central Anatolian Plateau. We used high-resolution topographic data, OSL ages, and published information on past sea levels to analyse the spatiotemporal evolution of these terraces; we derived a regional uplift model for the northward-advancing orogenic wedge that supports the notion of laterally variable uplift rates along the flanks of the Pontides. The best-fit uplift model defines a constant long-term uplift rate of 0.28 ± 0.07 m/ka for the last 545 ka. This model explains the evolution of the terrace sequence in light of active tectonic processes and superposed cycles of climate-controlled sea-level change. Our new data reveal regional uplift characteristics that are comparable to the inner sectors of the Central Pontides; accordingly, the rate of uplift diminishes with increasing distance from the main strand of the restraining bend of the North Anatolian Fault Zone (NAFZ). This spatial relationship between the regional impact of the restraining bend of the NAFZ and uplift of the Pontide wedge thus suggests a strong link between the activity of the NAFZ, deformation and uplift in the Pontide orogenic wedge, and the sustained lateral growth of the Central Anatolian Plateau flank.

An enhanced ensemble classifier for telecom churn prediction using cost based uplift modelling

Telecom, being a dynamic and competitive industry which contains an inherently high potential for customer churn, necessitating of accurate churn prediction models. Regular classification approaches fail to effectively predict churn due to low correlation levels between conventional performance metrics and business goals. This work presents an ensemble stacking incorporated with uplifting-based strategies for telecom churn prediction model. Evaluations have been performed based on conventional performance and a cost heuristic, with a major focus upon the cost heuristic. This mode of operation exhibits a high correlation levels between performance indicators and business goals, thus enabling the algorithm suitable for most cost-sensitive applications. A heterogeneous ensemble is created by using multiple algorithms to provide first level predictions. Those predictions with discrepancies are processed at the secondary level using a heuristic based combiner to provide the final predictions. Combination heuristics are fine-tuned based on the cost to predict more accurately concentrating on business goals. Subsequently, Customer uplifting is performed on final predictions, thus making the proposed model 50% more cost efficient than the state-of-the-art ensemble models.

Synthesis of the special issue: The formation and evolution of Ceres’ Occator crater

The distinctive bright regions within Occator crater are one of the most remarkable discoveries of the Dawn mission's exploration of Ceres. The central region is named Cerealia Facula and the additional regions in the eastern crater floor are named Vinalia Faculae. Here we summarize and synthesize the results of this special issue, which aimed to identify the driving forces behind the formation of Occator and the faculae, and thus lead us to a new understanding of the processes and conditions that occurred in Ceres’ past, and potentially in its present. The investigations presented here used Dawn data, theoretical modeling and laboratory experiments to deduce the sequence of events that led to the formation of Occator, Cerealia Facula and Vinalia Faculae, which are broken into stages 1–3. Stage 1: Occator's ejecta blanket, terraces and hummocky crater floor material formed during and shortly after crater formation. These features are located in many Cerean complex craters. However, Occator also contains the lobate materials and faculae, which are unique to Occator. We interpret the lobate materials as a slurry of water, soluble salts and boulders of unmelted silicates/salts, which flowed around the crater interior before solidifying. At least portions of the lobate materials were solidified prior to the formation of the central pit, which is suggested to form via the ‘melted uplift model’ and provides insights into central pit formation across the Solar System. We propose the outer edge of Cerealia Facula formed shortly after the Occator-forming impact, via impact-induced hydrothermal brine deposition or via salt-rich water fountaining (perhaps sourced in a pre-existing reservoir). Stage 2: the majority of Cerealia Facula is located within the central pit and is interpreted to have formed later, at least ∼18 Myr after the Occator-forming impact, via multiple depositional events. The Cerealia-Facula-forming brines may have flowed out of fractures in the walls of the central pit and/or been driven to the surface by freezing of a subsurface reservoir and/or deposited via salt-rich water fountains. Further investigations are required to identify whether the formation of the majority of Cerealia Facula is driven by processes triggered by an impact, i.e. an exogenic event, or by a combination of impact-driven and endogenic processes. Cryomagmatic intrusions are suggested to uplift the crater floor, resulting in concentric floor fractures and an asymmetric dome. Injection of a similar material is proposed to inflate part of the lobate materials, giving them a hummocky texture. The resulting stresses formed fractures in the hummocky lobate material, which allowed the Vinalia-Faculae-forming brines to travel to the surface, where they ballistically erupted. The central dome within the central pit was one of the last features to form, by laccolithic intrusion, or by volume expansion from freezing of volatiles, or by extrusion of brines. Stage 3: mixing with Ceres’ average materials and/or space weathering darken the faculae over time. Cerealia Facula and Vinalia Faculae are the brightest and freshest of the bright regions identified on Ceres’ surface. Bright regions darken over time until their eventual erasure. Thus, it is likely that faculae formation has occurred throughout Ceres’ history, but that Occator's faculae are visible today because they are geologically young. Through synthesis of the studies presented in this special issue, we find that entirely exogenic driving forces, triggered by the impact, or a combination of endogenic and impact-derived forces could explain the formation of Occator and its faculae. Whether activity is impact-triggered and/or endogenic in nature is a key question for all investigations of Ceres, and future studies may favor one possibility over the other. The investigations presented in our special issue indicate Ceres is an active world where brines have been mobile in the geologically recent past. As such, Ceres is an intriguing world that we have only begun to explore.

The central pit and dome at Cerealia Facula bright deposit and floor deposits in Occator crater, Ceres: Morphology, comparisons and formation

The prominent bright deposit Cerealia Facula, Ceres, coincides with the central depression (or central pit) of the recently formed 92 km-wide complex crater Occator. The central pit is 9–10 km wide and up to 1 km deep and is partially filled with a 700 m-high 2 km-wide dome. The upper surface of the central dome is densely fractured but the flanks are not, indicating that uplift of the dome surface occurred after the bright deposit was emplaced, and primarily through uplift of the surface from below by laccolithic intrusion or volume expansion. The pit is rimless except for two prominent massifs that flank it to the east and west. This pit-dome morphology bears a strong resemblance in morphology, topography, and dimensions to central pit craters observed on Ganymede and Callisto, but is essentially absent on the similar sized ice-rich moons of Uranus, Saturn, and Pluto. The lack of pit craters on midsize icy moons and the possible lack of central pit craters at the cold poles of Ceres suggest that temperature is a significant controlling factor in central pit formation, beyond the canonical inverse-gravity scaling of complex crater transitions. While renewed modeling will benefit from these new constraints, the mapping of central pits on Ceres, including Occator, suggests that either the rheologically layered target model or more likely the melted central uplift model may be the most probable explanation for central pit formation on Ceres (lack of similar high-resolution imaging on the Galilean icy satellites precludes definitive evaluation of models there). Large volumes of lobate floor-fill material at least 600 m thick in some places cover large areas of the southern and eastern floor of Occator, and small outcrops of this unit are found perched high on closed topographic depressions on the Occator terrace zone and other higher areas of the crater floor. The swirling and knobby surface textures of these lobate deposits are essentially indistinguishable from those on impact melt sheets on large lunar craters such as Tycho. These observations indicate that this unit is most likely an impact melt sheet formed from melted water ice admixed with unmelted particulate or dissolved phyllosilicates, salts and/or carbonates, and large quantities of fragmented debris. Melt of ice phases within Ceres crust is also supported by independent numerical models of impact at Occator and would provide a ready source of fluids to promote hydrothermal deposition of the observed bright Na-carbonates on the surface of the central pit at Occator, if the deposit formed within a million years or so of impact. The central dome finds analogs in salt domes on Earth and pingos (frost-mounds) on Earth and Mars, which form by water and solute migration and freeze expansion. Later freezing of a central reservoir of volatiles beneath the central pit at Occator or mobilization of ice or other phases during the cooling and refreezing process could explain the uplift of the central dome. Finally, no changes were detected in the Cerealia and Vinalia Facula bright deposits during the course of the prime Dawn mapping mission, placing additional constraints on their formation time and limits on ongoing activity.

Linear regression for uplift modeling

The purpose of statistical modeling is to select targets for some action, such as a medical treatment or a marketing campaign. Unfortunately, classical machine learning algorithms are not well suited to this task since they predict the results after the action, and not its causal impact. The answer to this problem is uplift modeling, which, in addition to the usual training set containing objects on which the action was taken, uses an additional control group of objects not subjected to it. The predicted true effect of the action on a given individual is modeled as the difference between responses in both groups. This paper analyzes two uplift modeling approaches to linear regression, one based on the use of two separate models and the other based on target variable transformation. Adapting the second estimator to the problem of regression is one of the contributions of the paper. We identify the situations when each model performs best and, contrary to several claims in the literature, show that the double model approach has favorable theoretical properties and often performs well in practice. Finally, based on our analysis we propose a third model which combines the benefits of both approaches and seems to be the model of choice for uplift linear regression. Experimental analysis confirms our theoretical results on both simulated and real data, clearly demonstrating good performance of the double model and the advantages of the proposed approach.

Predicting the Ideal Patient for Ambulatory Cleft Lip Repair.

The utilization of ambulatory surgical centers (ASCs) for cleft lip repair is increasing to reduce costs. This study better defines the patient population appropriate for ambulatory cleft repair with uplift modeling, a predictive analytics technique. Pediatric patients who underwent cleft lip repair were identified in the 2007 to 2011 California Healthcare Cost and Utilization Project State Inpatient Database and State Ambulatory Surgery and Services Database. The 2-model uplift approach was utilized using multivariate logistic regressions fit to assess the effect of ASCs, age, comorbidities, and procedure type on mortality or 30-day readmission. Of the pediatric cleft lip repairs in California between 2007 and 2011, 2383 (83%) were conducted in inpatient facilities and 498 (17%) in ASCs. The 30-day readmission rates were 2.01% and 1.93% for ASC repairs and inpatient repairs, respectively ( P = .909). Uplift modeling predicts that of the 2881 patients, approximately 40% of patients would have benefit from an ASC repair and an ASC repair would have had no effect on the remaining 60%. Patients likely to benefit from an ASC repair were more likely younger than 1 year old, nonsyndromic, not to have a respiratory or neurologic diagnosis, have less number of procedures, and to have undergone an isolated cleft lip repair. Uplift modeling predicts that approximately 40% of patients would benefit from an ASC cleft lip repair. Targeting patients younger than 1 year old, nonsyndromic, with no respiratory or neurologic diagnosis for ASC cleft lip repair, may be a safe and cost-saving endeavor.

Decomposition of Relative Sea Level Variations at Tide Gauges Using Results From Four Estonian Precise Levelings and Uplift Models

Variations in Estonian sea level records (1842–2016) are interpreted using harmonic and sliding window trend analysis together with the latest findings on the crustal movements in the region of the Baltic Sea. Considering the most recent relative-to-geoid land uplift (LU) rates from the semiempirical LU model NKG2016LU_lev, it was possible, probably for the first time, to properly interpret the climate-related sea level variation components in the Estonian tide gauges. The LU relative to geoid in Estonia (0.3–3.4 mm/yr) partly compensates for the global sea level rise (GSLR). Trends in Estonian relative sea level vary site dependently between –1.05 and 1.14 mm/yr (in 1954–2016) or between –0.17 and 1.14 mm/yr (in 1900–2010), yielding a “local sea level rise” (approximately 2.2 to 2.8 mm/yr in 1900–2010), which is faster by ∼0.8 mm/yr than the GSLR over similar period. This gap mainly originates from local land subsidence and variations in water exchange processes between the Baltic Sea and the World Ocean. Moreover, due to high intra- and inter-annual variability (standard deviations in detrended monthly data series from 20 to 22 cm), the sea level trend estimates are not stable enough and the usability of Estonian time series is relatively low for contributing to GSLR estimations.

Late Cretaceous to early Eocene deformation in the northern Tibetan Plateau: Detrital apatite fission track evidence from northern Qaidam basin

Unraveling the growth of northern Tibet is crucial to understanding the geodynamic processes of the India-Eurasia collision, evaluating plateau uplift models and reconstructing associated paleoclimate history. However, pre-middle Miocene deformation history of northern Tibet remains poorly understood. We use detrital apatite fission track (detrital AFT) thermochronology of Mesozoic and Cenozoic sedimentary rocks from northern Qaidam basin to constrain the early growth of northern Tibet. Detrital AFT ages of the Mesozoic samples are younger than their depositional ages, indicating that the Mesozoic succession underwent two stages of exhumation after deep burial during 80–61 Ma and 54–47 Ma. Detrital AFT ages of the Cenozoic samples are older than their depositional ages and reveal that their source region experienced two periods of exhumation (peak ages in 86–59 Ma and 54–36 Ma). These results suggest that the northern Tibet successively experienced Late Cretaceous–early Paleocene and early–middle Eocene deformation. The Late Cretaceous–early Paleocene deformation implies that the extent of pre-collisional (India-Eurasia) deformation region in the plateau were much larger than previously known. The static detrital AFT peak ages (54–51 Ma) based on lag-time analysis for the Paleogene samples demonstrate the early Eocene deformation was a rapid and short-lived event, which was a far-field response to the India-Eurasia collision. Hence, we advocate synchronous deformation throughout the northern plateau at the collision time. Lag-time analysis results also demonstrate absence of Oligocene to early Miocene cooling ages and post-Eocene decreasing exhumation rates in the Cenozoic source regions. This suggests relatively quiescent tectonic settings in the Qilian Mountains during the Oligocene–early Miocene and agrees that the lateral strike-slip movement of the Altyn Tagh Fault was accommodated out of the plateau by the end of the early Miocene.

Identification of Patients Expected to Benefit from Electronic Alerts for Acute Kidney Injury.

Electronic alerts for heterogenous conditions such as AKI may not provide benefit for all eligible patients and can lead to alert fatigue, suggesting that personalized alert targeting may be useful. Uplift-based alert targeting may be superior to purely prognostic-targeting of interventions because uplift models assess marginal treatment effect rather than likelihood of outcome. This is a secondary analysis of a clinical trial of 2278 adult patients with AKI randomized to an automated, electronic alert system versus usual care. We used three uplift algorithms and one purely prognostic algorithm, trained in 70% of the data, and evaluated the effect of targeting alerts to patients with higher scores in the held-out 30% of the data. The performance of the targeting strategy was assessed as the interaction between the model prediction of likelihood to benefit from alerts and randomization status. The outcome of interest was maximum relative change in creatinine from the time of randomization to 3 days after randomization. The three uplift score algorithms all gave rise to a significant interaction term, suggesting that a strategy of targeting individuals with higher uplift scores would lead to a beneficial effect of AKI alerting, in contrast to the null effect seen in the overall study. The prognostic model did not successfully stratify patients with regards to benefit of the intervention. Among individuals in the high uplift group, alerting was associated with a median reduction in change in creatinine of -5.3% (P=0.03). In the low uplift group, alerting was associated with a median increase in change in creatinine of +5.3% (P=0.005). Older individuals, women, and those with a lower randomization creatinine were more likely to receive high uplift scores, suggesting that alerts may benefit those with more slowly developing AKI. Uplift modeling, which accounts for treatment effect, can successfully target electronic alerts for AKI to those most likely to benefit, whereas purely prognostic targeting cannot.

Spatial and temporal variation of tectonic uplift in the southeastern Ethiopian Plateau from morphotectonic analysis

We use morphotectonic analysis to study the tectonic uplift history of the southeastern Ethiopian Plateau (SEEP). Based on studies conducted on the Northwestern Ethiopian Plateau, steady-state and pulsed tectonic uplift models were proposed to explain the growth of the plateau since ~30 Ma. We test these two models for the largely unknown SEEP. We present the first quantitative morphotectonic study of the SEEP. First, in order to infer the spatial distribution of the tectonic uplift rates, we extract geomorphic proxies including normalized steepness index ksn, hypsometric integral HI, and chi integral χ from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) digital elevation model (DEM). Second, we compare these rates with the thickness of flood basalt that we estimated from geological maps. Third, to constrain the timing of regional tectonic uplift, we develop a knickpoint celerity model. Fourth, we compare our results to those from the Northwestern Ethiopian Plateau to suggest a possible mechanism to explain regional tectonic uplift of the entire Ethiopian Plateau. We find an increase in tectonic uplift rates from the southeastern escarpments of the Afar Depression in the northeast to that of the Main Ethiopian Rift to the southwest. We identify three regional tectonic uplift events at ~11.7, ~6.5, and ~4.5 Ma recorded by the development of regionally distributed knickpoints. This is in good agreement with ages of tectonic uplift events reported from the Northwestern Ethiopian Plateau.

A Literature Survey and Experimental Evaluation of the State-of-the-Art in Uplift Modeling: A Stepping Stone Toward the Development of Prescriptive Analytics.

Prescriptive analytics extends on predictive analytics by allowing to estimate an outcome in function of control variables, allowing as such to establish the required level of control variables for realizing a desired outcome. Uplift modeling is at the heart of prescriptive analytics and aims at estimating the net difference in an outcome resulting from a specific action or treatment that is applied. In this article, a structured and detailed literature survey on uplift modeling is provided by identifying and contrasting various groups of approaches. In addition, evaluation metrics for assessing the performance of uplift models are reviewed. An experimental evaluation on four real-world data sets provides further insight into their use. Uplift random forests are found to be consistently among the best performing techniques in terms of the Qini and Gini measures, although considerable variability in performance across the various data sets of the experiments is observed. In addition, uplift models are frequently observed to be unstable and display a strong variability in terms of performance across different folds in the cross-validation experimental setup. This potentially threatens their actual use for business applications. Moreover, it is found that the available evaluation metrics do not provide an intuitively understandable indication of the actual use and performance of a model. Specifically, existing evaluation metrics do not facilitate a comparison of uplift models and predictive models and evaluate performance either at an arbitrary cutoff or over the full spectrum of potential cutoffs. In conclusion, we highlight the instability of uplift models and the need for an application-oriented approach to assess uplift models as prime topics for further research.

Fulfillment by Amazon Versus Fulfillment by Seller: A Risk-Adjusted Uplift Model

With dual-channel choices, E-retailers fulfill their demands by either the inventory stored in third-party distribution centers, or by in-house inventory. In this paper, using data from a wedding gown E-retailer in China, we analyze the differences between two fulfillment choices --- Fulfillment by Amazon (FBA) and Fulfillment by Seller (FBS). In particular, we want to understand the impact of FBA that will bring to sales and profit, compared to FBS, and how the impact is related to product features such as sizes and colors. We develop a risk-adjusted fulfillment model to address this problem, where the e-retailer's risk attitude to FBA is incorporated. We denote the profit gaps between FBA and FBS as the rewards for this E-retailer fulfilling products using FBA, our goal is to maximize the E-retailer's total rewards using predictive analytics. We adopt the Generalized Linear Model to predict the expected rewards, while controlling for the variability of the reward distribution. We apply our model on a set of real data, and develop an explicit decision rule that can be easily implemented in practice. The numerical experiments show that our interpretable decision rule can improve the E-retailer's total rewards by more than 35%.

A macro-element for the modelling of shallow foundation deformations under seismic load

This paper presents a new soil-foundation macro-element model to allow efficient and sufficiently accurate consideration of soil-foundation-structure interaction in structural analysis. The model makes use of two constitutive models, a plasticity model which models the soil inelastic deformation, and an elastic uplift model, which captures the geometric non-linearity during uplift of the foundation. Further considerations are made to allow the macro-element to be efficiently implemented in a particular non-linear finite element software (Ruaumoko3D). Two experimental centrifuge tests are simulated using the proposed macro-element: one of a bridge pier and one of a one-bay, one-storey frame structure. The simulated results are compared to the experimental behaviour to demonstrate the accuracy of the numerical model.

Onset Timing and Slip History of the Teton Fault, Wyoming: A Multidisciplinary Reevaluation

AbstractThe dramatic relief of the Teton Range in northwestern Wyoming results from motion along the Teton normal fault. New apatite (U‐Th)/He (AHe) and fission track (AFT) ages of samples collected in the footwall yield fundamental constraints on contrasting models of Teton fault activity and consequent relative footwall uplift. Low‐elevation samples in the immediate footwall of the fault range from 12.5 to 6.5 Ma. AHe ages of samples from subvertical transects range from 57.8 to 6.5 Ma (Rendezvous), 54.0 to 6.8 Ma (Grand), and 20.5 to 7.1 Ma (Moran), with all three transects yielding an expected trend of decreasing age with decreasing elevation. AFT ages obtained from the three transects range from 55.4 to 45.7 Ma (Rendezvous), 43.2 to 13.3 Ma (Grand), and 38.0 to 11.7 Ma (Moran). Inverse thermal history modeling indicates that the onset of relatively rapid cooling, as a proxy for relative footwall uplift, initiated first in the northern part of the range (15–13 Ma) and then migrated south as fault displacement continued (7 Ma to present at Rendezvous). Thus, this work suggests that although the Teton fault likely initiated due to Basin and Range extension, this structure remains active. Normal fault displacement models also suggest a minimum of ~6 km displacement for footwall uplift of ~2 km. If evaluated in the context of established normal fault length‐displacement relationships, this yields a length of ~180 km, substantially longer than current estimates of Teton fault length and putting the northern end of the fault well into, and potentially beyond, the bounds of Yellowstone National Park.