1) Bayesian hierarchical modeling for spatio-temporal crime incidence in urban districts 2) Forecasting electricity demand using state-space and machine learning hybrid models 3) Robust statistical methods for high-dimensional genomic data analysis 4) Causal inference with observational data: propensity score and double/debiased machine learning 5) Time series analysis of climate variables using irregularly spaced data 6) Nonparametric regression in survival analysis with censored data 7) Change-point detection in environmental and financial time series 8) Multivariate distribution modeling for extreme value analysis in risk assessment 9) Bayesian network structure learning for epidemiological data 10) Statistical methods for fair and interpretable predictive modeling in education analytics

Project Abstract

This study develops a comprehensive statistical framework integrating Bayesian hierarchical modeling, state-space forecasting, robust high-dimensional techniques, causal inference, irregular time-series analysis, nonparametric regression, change-point detection, extreme value modeling, Bayesian network structure learning, and fair interpretable predictive methods to address interconnected challenges across urban crime, energy demand, genomics, epidemiology, climate, survival analysis, finance, risk assessment, and education analytics. We propose a spatio-temporal Bayesian hierarchical model to quantify crime incidence in urban districts, incorporating space-time random effects, covariate information, and neighborhood adjacency structures to capture cross-district spillovers and nonstationarity. Simultaneously, we implement a hybrid forecasting architecture for electricity demand that blends state-space representations with machine learning components to exploit linear dynamics and nonlinear patterns, seasonal effects, and regime shifts, while quantifying predictive intervals through probabilistic forecasting. For high-dimensional genomic data, we develop robust statistical procedures—penalized and robustified estimators, stability selection, and false discovery rate control—designed to perform accurate variable selection and inference in the presence of multicollinearity and strong correlations. Causal inference is advanced via propensity score methods and double/debiased machine learning to estimate heterogeneous treatment effects from observational data, enabling valid policy-relevant inferences while accounting for model misspecification and confounding. Time series analysis of irregular climate data is addressed with irregularly spaced state-space models and ensemble approaches to ensure robust trend, seasonality, and anomaly detection under missingness and sampling gaps. Nonparametric regression is applied to survival data with censoring, enabling flexible modeling of covariate effects without rigid functional form assumptions, while change-point detection identifies structural shifts in environmental and financial time series, informing risk management and policy adaptation. Multivariate extreme value modeling is employed to characterize joint tail behavior for risk assessment across sectors, incorporating dependence structures relevant to concurrent extreme events. Bayesian network structure learning uncovers causal and probabilistic dependencies in epidemiological data, facilitating understanding of transmission dynamics and intervention impacts. Finally, fair and interpretable predictive modeling is pursued through techniques that emphasize transparency, feature importance, and bias mitigation in education analytics, ensuring that models support evidence-based decision-making while safeguarding equity. The integrated framework is validated on simulated benchmarks and applied to real-world datasets spanning crime, energy demand, genomics, health, climate, finance, and education datasets. Model diagnostics cover calibration, coverage, discrimination metrics, and sensitivity to prior choices, with emphasis on interpretability and policy relevance. The study contributes a unified methodology for simultaneous inference, forecasting accuracy, causal assessment, and fairness in complex, heterogeneous data environments, offering practical guidelines for researchers and practitioners in statistics, data science, and public policy.

Project Overview

What This Project Is About

---

The project explores a range of modern statistical methods and how they help analyze real-world data. From crime patterns in cities to predicting electricity use, and from genomic data to weather signals, it links ideas in statistics to practical problems. The goal is to build approachable explanations and simple models that a final-year undergraduate can implement with basic tools.

---

---

The Problem It Addresses

---

Many important data problems involve uncertainty, complex patterns over space and time, and limited or irregular data. Students learn to recognize these challenges and to choose methods that are robust, interpretable, and useful for decision-making in policy, business, health, and the environment.

---

---

Objectives of the Project

---

1. Describe the core ideas behind each topic in plain language.
2. Identify real-world problems where these methods could help.
3. Develop a small, workable workflow or prototype for one topic of interest.
4. Explain limitations and assumptions of the methods used.

---

---

What You Will Do Step by Step

---

1. Choose one or two topics to focus on and summarize the data needs.
2. Find or simulate simple datasets that illustrate the method.
3. Demonstrate the method with beginner-friendly software (e.g., spreadsheets or basic programming).
4. Interpret results in clear terms, highlighting what decisions they inform.
5. Prepare a short report and a one-page summary for non-experts.

---

---

Expected Outcome

A concise understanding of several statistical approaches, plus a ready-to-run example or plan for a deeper project. Learners should be able to explain when to use each method and how it can improve real-world analysis.