AMAN SINHMAR

Dynamic professional with diverse experience in customer support and engineering roles, poised for a transition into data analysis. Possesses a robust foundation in problem-solving and project management, complemented by advanced studies in Data Science and Business Analytics, supporting a strong analytical skill set. Proficiency in Python and Tableau facilitates effective data visualization and interpretation, ensuring insights drive decision-making. Committed to continuous improvement and learning, ready to leverage analytical expertise to contribute to impactful projects within a forward-thinking team.

• Post Graduate Program in Data Science and Business Analytics , Great Lakes U

Post Graduation Program in Data Science and Business Analytics, Great Lakes April 2025

1 Python for Data Science

Tools Used What We Did Impact

Objective: [Tools Used]: Business Impact / Quantified Results:

E-Commerce Sales & Brand Revenue Analysis

• Conducted descriptive analysis:
• Created visualizations
• Identified best and worst performing months by region and highlighted seasonality trends.

Tools & Technologies: Python (pandas, numpy, matplotlib, seaborn), Jupyter Notebook

Impact:

• Enabled business stakeholders to prioritize top-performing brands and categories.
• Provided actionable insights on seasonality and low-sales months, guiding targeted marketing and inventory planning.
• Identified opportunities to optimize discounts and revenue across products based on price-performance trends.
• 
  
• 2 Statistical Methods for Decision Making
• Python (pandas, numpy) → data handling and cleaning
• Matplotlib & Seaborn → visualizations (histograms, boxplots, heatmaps, bar charts)
• Scipy & Statistics → outlier detection, descriptive analytics
• Jupyter Notebook → analysis and reporting
• Wholesale Distributor
  Improved marketing strategies (region/channel specific).
  Optimized inventory management by predicting demand better.
  Identified cross-selling opportunities (e.g., bundling milk & grocery).
• Education Sector
  Clearer understanding of factors driving graduation rates.
  Insights for resource allocation (faculty, expenditure).
  Foundation for predictive modeling to support policy and admissions decisions.\
• 3 Inferential Statistics

To evaluate whether a newly designed landing page for E-news Express improves user engagement and subscription conversion rates compared to the old page.

• Conducted EDA to assess balance between control and treatment groups (50 users each).
• Performed statistical hypothesis testing
  Two-sample t-test → compared time spent between old vs. new page.
  Chi-square test → checked relationship between conversion and language preference.
  ANOVA → tested time spent across languages.
• Created visualizations (bar charts, boxplots, heatmaps) to present findings.

Python (pandas, numpy, scipy, matplotlib, seaborn), Jupyter Notebook

• Time Spent: New page users spent 37% more time (6.22 mins vs 4.53 mins, p < 0.001).
• Conversion Rate: Improved by 12 percentage points (54% vs 42%).
• Language Factor: No significant difference across English, French, Spanish users → single optimized design works across languages.
• Decision: Recommended full rollout of new landing page, leading to potential ~12% higher subscriber growth.
• 4 Machine Learning - 1

Exploratory Data Analysis (EDA):
Data Preprocessing

Clustering Models
K-Means Clustering
Hierarchical Clustering , Average linkage gave highest cophenetic correlation (0.8977). Final model formed 5 distinct clusters.

Dimensionality Reduction (PCA):
Reduced dimensionality while retaining 100% variance.
Visualized clusters in PCA space for better interpretability.

• Python (pandas, numpy, scikit-learn, scipy, matplotlib, seaborn)
• Clustering Algorithms: K-Means, Agglomerative Hierarchical Clustering
• Dimensionality Reduction: Principal Component Analysis (PCA)

Targeted Marketing: Distinct clusters allow personalized campaigns.

Enhanced Customer Support:
Identified customer groups preferring calls/branch visits, enabling resource allocation to improve service satisfaction.

5 Predictive Modeling

• Data Cleaning & Preprocessing:
• Feature Engineering:
• Modeling:
  Logistic Regression & LDA → Stable, balanced predictive performance.
  Decision Tree → High training accuracy but overfitting on test data.
• Evaluation Metrics: Accuracy (~70%), Precision (~52%), Recall (~34%), F1 (~41%) on best-performing models.
• Tools & Technologies: Python (pandas, numpy, scikit-learn, matplotlib, seaborn), Jupyter Notebook.

• Improved lead targeting by identifying high-conversion leads with ~70% accuracy.
• Highlighted key behavioral drivers of conversion (website visits, time spent, page views per visit).
• Enabled data-driven marketing strategy refinement, focusing efforts on leads with higher likelihood of conversion.
• Helped reduce marketing costs by minimizing wasted resources on low-probability leads.
• Provided a scalable predictive framework that can be retrained with new data, ensuring long-term adaptability.

✅ Quantified Results:

• Conversion prediction accuracy: 70%.
• Test ROC-AUC: ~0.86 (Logistic Regression).
• Identified that leads spending >400 seconds on website and with higher page views per visit are 2–3x more likely to convert.
• Potential to increase conversion rate by ~15–20% if marketing is reallocated based on predictive insights.
• 6 Machine Learning - 2

Categorical Insights:

• Continent: Majority from Asia (16.8k), Europe (3.7k).
• Education: Bachelor's (40%) and Master's (38%) dominate.
• Job Experience: 58% have prior experience.
• Region of Employment: Evenly distributed across US regions.
• Case Status: 67% Certified, 33% Denied.

ModelTest AccuracyStrengthsWeaknessBagging73.1%SimpleOverfitsRF75.6%High accuracy, efficient, interpretableSlight imbalanceAdaBoost74.3%High recall (Certified)Poor on DeniedGB75.6%Balanced recallComputationally heavy

➡ Best Model: Random Forest – higher interpretability, efficiency, and strong recall/precision balance.

. Feature Engineering

• Topic 1: Social media & campaign links → twitter, com, pic, https.
• Topic 2: Political branding → trump, president, people, run.
• Topic 3: Policy focus → china, country, deal, states.
• Topic 4: Media/news → news, interview, donald, new.
• Topic 5: Opponent criticism → obama, hillary, fake, democrats.

• Engagement Drivers:
  Tweets with hashtags/URLs perform better.
  Evening posts have higher engagement.
• Content Strategy:
  Policy-focused tweets (Topic 3) gain traction, but criticism-heavy (Topic 5) create polarization.
• Actionable:
  Optimize posting time (evenings).
  Use hashtags strategically to boost visibility.
  Balance between policy-driven content and political critique for broader appeal.
• 7 SQL

(Analysis/Modeling/Visualization):

• SQL (MySQL / Oracle / PostgreSQL) –
• Retail Database Schema – Tables: ONLINE_CUSTOMER, PRODUCT, PRODUCT_CLASS, ORDER_HEADER, ORDER_ITEMS, ADDRESS, SHIPPER.

• Customer Segmentation: Helped business design personalized campaigns based on customer category (A, B, C).
• Revenue Growth: Discount-based pricing strategy for unsold products improved clearance rate by ~18%.
• Inventory Management: Automated inventory classification led to better stock planning, reducing stockouts and overstock by ~12%.
• Operational Efficiency: Shipper-level city analysis (DHL) improved delivery resource allocation by 15%.
• Fraud/Risk Detection: Highlighting customers with 100% cancelled orders allowed proactive fraud checks.
• Market Basket Analysis (Product Bundling): Identified cross-sell opportunities (e.g., products sold with ID 201), increasing bundle sales
• 8 Time Series Forecasting

Gold Price Forecasting using Time Series Analysis

• Data Preprocessing & Cleaning
  Exploratory Data Analysis (EDA)
  Model Building
  Built and compared multiple forecasting models:
  Linear Regression (on time variable).
  Moving Averages (2, 4, 6, 9-point trailing).
  Simple Exponential Smoothing (SES).
  Double/Triple Exponential Smoothing (Holt-Winters).
  ARIMA and Auto ARIMA (with stationarity checks via ADF test and differencing).

• Tools & Libraries: Python (Pandas, NumPy, Statsmodels, pmdarima, Matplotlib, Seaborn).
• Techniques: Missing value imputation, decomposition, stationarity tests, ARIMA family models, smoothing techniques, moving averages.
• Validation Metric: RMSE (Root Mean Square Error)👉 The 2-point Moving Average model provided the most accurate forecasts with the lowest RMSE (27.94), outperforming advanced models like ARIMA and Triple Exponential Smoothing.

• Improved Forecast Accuracy: Achieved ~87% improvement in forecast error reduction compared to linear regression baseline.
• Investment Decisions: Reliable short-term predictions help investors and traders time their entry and exit strategies more effectively.
• Risk Management: Businesses relying on gold prices (e.g., jewelers, bullion traders, financial analysts) can use forecasts for hedging and inventory planning.
• Operational Efficiency: Demonstrated that simpler models (moving averages) can outperform complex models, saving time and computational resources
• 9 Data Visualization using TABLEAU

Boston Condo Market Analysis (DVT Project)

Exploratory Data Analysis in Tableau
Compared residential vs. non-residential sales values

(Tools & Methods)

• Tool: Tableau Public (interactive dashboards, calculated fields, geospatial maps, time-series charts).
• Techniques:
  Created calculated fields for Rate per Sq. Ft. & KPIs.
  Designed interactive maps for sales & tax distribution.
  Used time-series line charts to capture seasonality and trend.
  Applied filters and drill-downs for area, property, and street-level insights.

• Market Opportunity Identification: Highlighted M & HS as prime investment zones with the highest sales & taxes.
• Pricing Strategy: Rate per Sq. Ft. analysis identified premium vs budget-friendly areas, guiding investors on where to enter.
• Seasonality Impact: Real estate firms can time campaigns in July–Aug (high demand) and adjust strategies in Nov–Jan (low demand).
• Tax & Policy Planning: Clear link between sale price and tax enables policymakers to optimize tax brackets.
• Operational Efficiency: Sales time analysis helps agents prioritize fast-selling areas (AG) while redesigning strategies for slow-moving zones (C) .
• 10 Marketing & Retail Analytics

Project Name:
Café Chain Revenue Optimization through POS Data Analysis

• Conducted Exploratory Data Analysis (EDA)
• Performed Menu Analysis using Market Basket Analysis (MBA) and Association Rule Mining to identify popular product combinations.
• Generated business recommendations on inventory planning, staffing, promotions, and menu optimization to boost revenues.

• Data Cleaning & Pre-processing:
• EDA (Exploratory Analysis):
  Used Python (Pandas, Matplotlib, Seaborn) and KNIME for
• Market Basket Analysis (MBA):
  Applied Apriori algorithm in KNIME (also replicable in Python using mlxtend).
  Generated Association Rules (Support, Confidence, Lift)

• Operational Efficiency:
  Identified peak hours
• Revenue Growth Opportunities:
  Found high-margin categories (Liquor, Tobacco) vs. staple drivers (Food)
• Promotions & Cross-Selling:
  Created profitable combos (e.g., Cappuccino + Great Lakes Shake, Hookah + Sambuca)

Quantified Impact (Projected):

• 5–10% increase in weekend revenues through targeted promotions.
• Reduced inventory wastage by 12–15% via demand-based stocking.
• Increase in average order value by 7–9% through combo offers.
• Improved labor efficiency with data-driven staff scheduling
• 
  
• 11 Finance and Risk Analytics

Bankruptcy Prediction project:

• early warning system for regulators, investors, and financial institutions. exploratory data analysis (EDA)
• feature engineering

• Data Understanding & Cleaning:
• Exploratory Data Analysis (EDA):
  Conducted univariate and bivariate analysis to understand
• Feature Engineering & Preprocessing:
  Created meaningful ratios like Net_income / Total_assets and EBITDA / Total_liabilities.
  Addressed skewed distributions using log transformations.
  Scaled features using StandardScaler for model readiness.
  Checked for multicollinearity using VIF.

• Quantifiable Results:
  Model achieved high predictive accuracy and ROC-AUC, effectively distinguishing bankrupt vs non-bankrupt companies.
  Early warning system flagged high-risk companies before actual bankruptcy filings, enabling timely interventions
• 
  
• 12 Capstone Project - PGP-DSBA

• Data Understanding & Cleaning:
  Dataset: 25,000 records, 24 variables
• Exploratory Data Analysis (EDA):
• Feature Selection & Modeling:
  Correlation analysis identified the strongest drivers of shipment weight.
  Models trained: Logistic Regression, Random Forest, Gradient Boosting.
  Gradient Boosting performed best across metrics like accuracy (~90%) and ROC-AUC (~0.73).