Kushagra is currently serving as a Senior Design for Supply Chain Engineer at Micron Technology. He work alongside product development teams to enhance supply chain efficiency through design commonality and modularization. His role involves applying advanced statistical and machine learning methods to optimize supply chain strategies and developing executive-level dashboards to communicate performance metrics, trade-offs, and key structural levers.
Research Interest
Social Network Analysis, Statistical Methods, Generative Modelling, Agent-Based Simulation and Causal Inference
Projects
Socio-Spatial Patterns of Suicide Mortality in the United States
- Integrated Facebook's Social Connectedness Index with county-level suicide mortality data (2010–2022) to quantify how suicide risk and protective policies propagate through inter-county social networks
- Implemented two-way fixed effects regression models with sociodemographic, economic, and spatial controls across county-year observations to disentangle social versus geographical proximity effects
- Demonstrated that a one-standard-deviation increase in social exposure to ERPO-implementing states is associated with a reduction of 0.214 suicide deaths per 100,000 people, even without local policy enactment
- Validated findings through robustness checks using age-adjusted mortality rates, spatial exposure controls, and state-by-year fixed effects, confirming persistence of both social influence and indirect ERPO effects
- Highlighted that prevention strategies targeting socially central "hub" counties can maximize indirect impact through faster diffusion of help-seeking norms and firearm safety practices across networks
Socio–Spatio–Temporal Predictions of Opioid Overdose Deaths in the United States Counties
- Developed an XGBoost forecasting framework using SCI-based social proximity and inverse-distance spatial proximity features to predict county-level opioid overdose mortality at 1- and 2-month horizons across US counties (2018–2021)
- Benchmarked three nested XGBoost specifications — Baseline (autoregressive lags + covariates), Baseline + Social Proximity, and Baseline + Spatial Proximity, alongside EWMA-smoothed two-way fixed effects models to evaluate socio-spatial predictive gains
- Demonstrated that social proximity yields the lowest out-of-sample RMSE across both forecast horizons, consistently outperforming baseline and spatial-only models
- Quantified temporal decay of socio-spatial influence using EWMA smoothing within a two-way fixed-effects design, identifying half-lives of ~9 days (social) and ~15 days (spatial) via a 2D grid search over decay parameters
- Showed that monthly lag-based models explain near-zero residual variation (R² = 0.004) compared to EWMA-encoded specifications (R² = 0.763), establishing that network-mediated excitation operates on sub monthly timescales
- Motivated a continuous-time socio-spatial Hawkes process formulation as a process-based extension, grounded in the empirically inferred short-memory dynamics and rapid cross-location triggering
Measuring Network Dynamics of Opioid Overdose Deaths in the US University of Pittsburgh, USA
- Proposed a novel method to investigate the statistically significant effect of social influence on opioid overdose deaths
- Utilized Facebook’s social connectedness index (SCI) to construct the metric for social influence implemented two stage-least squares, cluster-robust, two-way fixed effect, network, and spatial autocorrelation linear models
- Demonstrated that one standard deviation increase in social proximity leads to 13 more deaths per 100,000 population in the USMeasuring the Effect of Social Networks on Suicide Ideation in the US University of Pittsburgh, USA
- Showed network cohesiveness is negatively associated with suicide rates in the US counties using negative binomial regression
- Proposed the use of SCI to create a network measure to analyze the dynamics of suicide ideation in US counties
- Created a framework under a staggered difference-in-difference model using exogenous policy shock to estimate the network effect
Publication
Socio-Spatial Patterns of Suicide Mortality in the United States | medRxiv [link]
Tiwari, K., Rahimian, M.A., Roberts, M.S. et al. Measuring network dynamics of opioid overdose deaths in the United States. Sci Rep 14, 29563 (2024). [link]
Products
- Social Proximity using Facebook's Social Connectedness Data- Code
Poster Sessions
- Poster Competition INFORMS Annual Meeting 2022, Indianapolis: Analyzing the effect of social connections on Opioid Overdose Deaths.
Talk Sessions
- Modeling Social Contexts of Opioid Epidemics Using Aggregate Facebook Connectivity Data in a Large-scale Agent-based Model of the U.S. INFORMS Annual Meeting 2022, Indianapolis, 2022.
- Peer Influence and Spread of Opioid Epidemic: A Data Driven Social Network Analysis Approach Using Facebook’s Social Connectedness Index. INFORMS Annual Meeting 2023, Phoenix
- Measuring Network Dynamics of Drug Overdose Deaths in the US. INFORMS Annual Meeting 2024, Seattle, 2024
Education
B.Tech in Production and Industrial Engineering, Vellore Institute of Technology.Education
Internship
Internship at Merck Sharp & Dohme LLC
- Role: Associate Specialist
- Project: End-to-End Supply Chain Optimization with Advanced Machine Learning in Air-to-Ocean Logistics
Awards
(2022) Teaching Assistant of the Year
