Academic project portfolio

Featured Projects

Applied research and analytics work across network science, experimental analysis, regression modeling, supervised learning, unsupervised learning, computer vision, and business analytics.

6,429

Facebook friendship ties modeled

Copenhagen network analysis

40.8%

Facebook ties overlapping Bluetooth co-presence

Copenhagen network analysis

F = 14.64

Advisor-source ANOVA statistic

AI advice experiment

93%

Face-recognition accuracy after tuning

Independent CV project

20k+

Refurbished-device sales records analyzed

ReCell pricing model

p ≈ .074

Scheduling interaction signal

Cloud scheduling regression

Network Analysis2026

Physical Co-Presence, Communication, and Facebook Friendship

DACSS Research Project

A network-inference analysis of the Copenhagen Networks Study examining how Bluetooth co-presence, SMS communication, gender, triadic closure, and Facebook friendship are related.

Research Question
Do physical co-presence and deliberate communication independently predict explicit Facebook friendship ties after accounting for network structure?
834
Facebook nodes
6,429
Facebook ties
0.244
Global clustering
vs. random mean 0.018
40.8%
FB–Bluetooth overlap
3.25×
Bluetooth ERGM OR
111.8×
SMS ERGM OR
large; interpret cautiously
Methods
Descriptive network analysisERGMLouvain communitiesScaffolded SAOM framework
Findings
  • Facebook friendship was sparse but structurally clustered, with clustering far above random-graph expectation.
  • Bluetooth co-presence and SMS communication both showed positive associations with Facebook friendship in the ERGM.
Interpretation note: The gender term in the ERGM coefficient table is statistically significant but reported with a negative coefficient, so the direction should be described carefully rather than overstated as positive homophily.
RERGMRSienaNetwork statisticsCommunity detection
Experimental Analysis

AI Advice-Seeking Experiment

DACSS 602

A randomized survey experiment testing whether people judge advice-seeking differently when the advisor is an AI chatbot, a friend, or a therapist in an interpersonal conflict scenario.

Research Question
Does advisor source change perceived appropriateness of seeking relationship-conflict advice?
250
Sample size
84
AI chatbot n
85
Friend n
81
Therapist n
F = 14.64
ANOVA
9.8e-07
p-value
Methods
Survey experimentData cleaningOrdinal recodingOne-way ANOVA
Findings
  • Advisor source significantly affected appropriateness ratings.
  • Friends and therapists received higher mean appropriateness ratings than the AI chatbot condition.
RtidyverseANOVASurvey designData visualization
Regression AnalysisMay 2026

Scheduling Structure and Task Completion Time

Cloud Task Scheduling Dataset

A regression study of whether load imbalance and deadline pressure jointly shape task completion time in a cloud task scheduling system.

Research Question
Does the effect of load imbalance on completion time depend on deadline pressure?
5
Models estimated
M5
Primary model
≈ 0.074
Interaction p
standard SE; HC3 table is primary
~0.001–0.002
Adjusted R² range
HC3
Inference
Methods
Five linear modelsInteraction modelHC3 robust standard errorsResidual diagnostics
Findings
  • Main effects for load imbalance and deadline pressure were weak when evaluated in isolation.
  • The primary M5 interaction model produced suggestive evidence at the 90% confidence level that the imbalance effect depends on deadline context.
Interpretation note: The result is exploratory and conditional, not evidence that imbalance or deadlines independently drive completion time.
RLinear regressionModerationHC3 robust SEDiagnostics
Computer VisionIndependent

Face Recognition Software

A deep-learning system for real-time and static face detection, tuned for accuracy and faster inference.

93%
Accuracy
25%
Inference improvement
Methods
Computer vision pipelineHyperparameter tuningTensorFlow GPU acceleration
Findings
  • Model tuning improved classification performance.
  • GPU acceleration reduced inference time for deployment-oriented use cases.
PythonTensorFlowComputer VisionDeep Learning
Supervised Learning

Dynamic Pricing Model for ReCell

UT Austin / Great Learning

A regression-based pricing analysis of used and refurbished devices, focused on resale-value drivers and dynamic pricing strategy.

Research Question
Which device attributes meaningfully influence resale price, and how can they support dynamic pricing?
20k+
Dataset scale
refurbished-device sales records
Linear regression
Model family
Pricing rules
Output
Methods
EDALinear regressionAssumption checksBusiness recommendations
Findings
  • The project linked predictive modeling to pricing-policy decisions rather than treating accuracy as the only output.
  • Feature interpretation was used to identify factors that materially influence refurbished-device value.
PythonEDALinear RegressionBusiness Analytics
Unsupervised Learning

Stock Data Clustering / Trade&Ahead

UT Austin / Great Learning

An unsupervised analysis grouping stocks by provided attributes and surfacing cluster-level investment patterns.

Research Question
Can stocks be segmented into interpretable groups that support portfolio diversification decisions?
2
Algorithms
K-means + hierarchical
Diversification
Use case
Methods
EDAK-means clusteringHierarchical clusteringCluster profiling
Findings
  • Cluster profiling translated unsupervised model output into interpretable investment segments.
  • The project emphasized portfolio insight rather than only algorithmic separation.
PythonK-meansHierarchical ClusteringEDA
Model Tuning

ReneWind Generator Failure Prediction

Great Learning

A turbine-generator failure prediction project using sensor data to identify failures before breakdown and reduce maintenance cost.

Research Question
Which tuned classification model can best identify generator failure risk early enough to support preventive maintenance?
Failure classification
Modeling task
Preventive maintenance
Operational target
Methods
Classification modelsUp/down samplingRegularizationHyperparameter tuning
Findings
  • The project focused on cost-sensitive operational value: identifying failures before generator breakdown.
  • Sampling and tuning were central because failure prediction often involves imbalance and asymmetric costs.
PythonClassificationRegularizationHyperparameter Tuning
Ensemble Learning

EasyVisa Approval Predictive Model

Great Learning

A classification project predicting visa certification outcomes and identifying applicant-profile factors associated with approval decisions.

Research Question
Which applicant profiles are most suitable for certification or denial based on predictive model evidence?
3
Ensemble families
bagging, boosting, stacking
GridSearchCV
Tuning method
Methods
EDAData preprocessingBaggingBoostingStackingGridSearchCV
Findings
  • The analysis used ensemble methods to compare predictive performance and identify influential applicant features.
  • The business output was a profile recommendation framework, not just a binary prediction.
PythonRandom ForestAdaBoostGradient BoostingXGBoost
Classification

INN Hotels Booking Cancellation Model

Great Learning

A predictive classification analysis of hotel booking cancellations designed to support refund and cancellation-policy decisions.

Research Question
Which booking characteristics most strongly signal cancellation risk before arrival?
Logit + Tree
Models
AUC-ROC
Evaluation
Methods
EDAData preprocessingLogistic regressionDecision treeAUC-ROC
Findings
  • The project connected classification outputs to policy decisions around overbooking, refunds, and cancellation management.
  • Decision-tree pruning and AUC-ROC evaluation supported model selection and interpretability.
PythonClassificationLogistic RegressionDecision Tree
A/B Testing

E-news Express Landing Page Effectiveness

Great Learning

A statistical testing project evaluating whether a new landing page improved subscriber conversion and whether conversion depended on language preference.

Research Question
Is the new landing page effective enough to gather new subscribers, and does language preference matter?
A/B test
Design
Conversion + dwell time
Core outcomes
Methods
A/B testingHypothesis testingStatistical inferenceData visualization
Findings
  • The project used conversion status and time-on-page as decision metrics for product evaluation.
  • Segmentation by preferred language connected the experiment to user-experience decisions.
PythonA/B TestingHypothesis TestingVisualization
Data Analytics

FoodHub Order Analysis

Great Learning

An exploratory analytics project using food-order data from an aggregator platform to answer operational business questions.

Research Question
What order, customer, and operational patterns can inform business improvement?
EDA
Analysis type
Ops insights
Business output
Methods
Variable identificationUnivariate analysisBivariate analysisBusiness recommendations
Findings
  • The project emphasized actionable insights for improving food-aggregator operations.
  • Exploratory analysis was used as a decision-support workflow rather than as a purely descriptive exercise.
PythonEDAVisualizationBusiness Analytics
Results dashboard

Selected Quantitative Results

These visuals summarize interpretable results from the uploaded research reports rather than adding decorative charts.

Advisor Appropriateness Means

Mean appropriateness rating by advisor source on a 1–4 scale in the AI advice-seeking experiment.

AI Chatbot
n = 84
2.25
Friend
n = 85
2.88
Therapist
n = 81
2.90

ANOVA result: F(2, 247) = 14.64, p = 9.8e-07.

Facebook Friendship ERGM Odds Ratios

Selected positive ERGM effects from the Copenhagen network analysis. Larger odds ratios indicate stronger conditional association with Facebook friendship ties.

Bluetooth co-presence
3.25×
Triadic closure
16.32×
SMS communication
111.82×

The SMS odds ratio is very large and should be read as evidence of strong association, not a causal effect.

Behavioral Network Activity by Wave

Bluetooth co-presence and SMS volume both peaked in Wave 2, making it a useful diagnostic point for network activity rather than a decorative trend.

Wave 1dyadic logs
Bluetooth ties13,242
SMS messages6,366
Wave 2dyadic logs
Bluetooth ties20,336
SMS messages7,071
Wave 3dyadic logs
Bluetooth ties16,485
SMS messages5,595
Wave 4dyadic logs
Bluetooth ties16,378
SMS messages5,301
Great Learning Journey

PGP-BABI-INTL Project Milestones

A structured view of completed Great Learning projects, emphasizing objective, method, and applied value.

Project completed

ReCell

Supervised Learning — Foundations

Objective
Analyze a used-device dataset, build a model for dynamic pricing, and identify price-driving factors.
Applied value
Turns model coefficients into pricing strategy.
#EDA#LinearRegression#AssumptionChecks#BusinessRecommendations
Project completed

Trade&Ahead

Unsupervised Learning

Objective
Group stocks based on attributes and explain the characteristics of each group.
Applied value
Supports portfolio segmentation and diversification thinking.
#K-means#HierarchicalClustering#ClusterProfiling
Project completed

ReneWind

Model Tuning

Objective
Use sensor data to predict wind-turbine generator failure before breakdown.
Applied value
Links classification to preventive maintenance and cost reduction.
#Sampling#Regularization#HyperparameterTuning
Project completed

EasyVisa

Ensemble Techniques

Objective
Predict visa approval status and identify applicant profiles suitable for certification or denial.
Applied value
Uses ensemble learning for applicant-profile recommendations.
#Bagging#RandomForest#Boosting#Stacking#GridSearchCV
Project completed

INN Hotels

Supervised Learning — Classification

Objective
Predict booking cancellations and identify factors influencing cancellation risk.
Applied value
Connects prediction to refund and cancellation-policy design.
#LogisticRegression#DecisionTree#Pruning#AUC-ROC
Project completed

E-news Express

Business Statistics

Objective
Evaluate whether a new landing page improves subscriber conversion using experimental evidence.
Applied value
Turns simulated product data into a launch decision.
#HypothesisTesting#A/BTesting#Visualization#StatisticalInference
Project completed

FoodHub Order Analysis

Python — Foundations

Objective
Analyze food aggregator order data to answer operational business questions.
Applied value
Creates actionable operations recommendations from order data.
#Python#UnivariateAnalysis#BivariateAnalysis#EDA

Interested in Collaboration?

Open to projects involving data science, machine learning, behavioral analytics, network analysis, experimental design, and applied AI.

Let's Connect