Everything I've built, in detail.

An end-to-end statistical analysis of one month of FoodExpress transactional data that evaluates delivery performance, customer satisfaction drivers, cuisine-level behavior, and promotional impact. Purpose: To identify operational levers—especially delivery speed and customer segmentation—and recommend prioritized, actionable changes to improve on-time performance, retention, and revenue. Origin: Prepared as a reproducible Jupyter Notebook study by Sakib Ahmed (raw CSV → cleaning → EDA → hypothesis tests → visualizations), with outputs included as PNG figures and a PDF report.

A Jupyter Notebook pipeline ingests the raw CSV, applies cleaning and outlier handling, runs EDA and statistical tests (correlation matrices, t-tests, one‑way ANOVA, Q–Q and KDE checks), and produces visualizations and a PDF summary. Core tools: Python (Pandas, NumPy), Matplotlib/Seaborn for charts, and SciPy for hypothesis testing; final deliverables are notebook, PNG figures, and a PDF report.

• Reproducible notebook pipeline: data ingestion from foodexpress_data_uncleaned.csv, cleaning (missing/INVALID values and extreme outliers handled), and staged EDA enabling repeatable analyses and easy extension.
• Statistical rigor and diagnostics: computed correlations and significance tests (independent t-tests, paired t-tests, one‑way ANOVA) plus distribution checks (Q–Q plots, KDEs) to validate assumptions and back recommendations with p-values and effect sizes.
• Clear, data-driven visualizations: generated focused plots (delivery-time distributions, cuisine order volumes, delivery-time by zone, correlation heatmaps, rating distributions) and a sample dashboard image to communicate operational insights to stakeholders.
• Business-metric extraction: derived summary KPIs (mean/median delivery times, VIP share = 52.7%, net revenue) and reported a strong negative correlation between delivery time and customer rating (r = −0.46) as the primary actionable lever.
• Actionable deliverables & documentation: consolidated findings into a PDF report with prioritized recommendations (routing optimization, peak-hour staffing, VIP benefits, campaign replication) and included the dataset and figures for auditability and further modeling.

What: This is a focused R project that analyzes the built-in mtcars dataset to explore relationships between vehicle characteristics (e.g., mpg, horsepower, weight, cylinders). Purpose: It demonstrates an end-to-end small-scale analysis workflow — data cleaning, exploratory data analysis, visualizations, and basic statistical modeling (e.g., linear regression) — intended for learning, demonstration, or quick prototyping. Origin: Created as a compact, R-native example repository using R scripts and R Markdown to document methods and findings, making the work easy to reproduce and adapt.

The codebase is implemented entirely in R, organized as scripts and R Markdown documents that read the built-in mtcars data, perform tidyverse-based wrangling, generate ggplot2 visualizations, fit statistical models (base lm or modeling helpers), and render results as reproducible reports (HTML/PDF) and figure artifacts. The workflow is script-driven and designed for straightforward extension.

• Reproducible analysis: Clear, runnable R scripts / R Markdown that reproduce the full analysis and narrative results.
• Tidy data pipeline: Uses tidyverse idioms for concise data transformation and filtering, keeping transformations transparent and easy to audit.
• Informative visualizations: ggplot2-based plots that surface key patterns (mpg vs weight, horsepower effects, cylinder-group comparisons) to support conclusions.
• Statistical modeling & diagnostics: Includes simple regression models with interpretation-ready outputs and diagnostic checks to validate assumptions.
• Lightweight, educational design: Minimal dependencies and a compact code layout make the project easy to read, fork, and adapt for teaching or quick experiments.
• Extensibility: Structured so users can swap datasets, add models, or convert analysis into interactive formats (Shiny) or CI-driven report generation with minimal changes.

This project analyzes an A/B experiment that compares an incumbent e‑commerce landing page with a redesigned variant to measure conversion-rate differences.

It uses statistical testing and visualizations to answer whether the new page meaningfully improves conversions and to inform a rollout decision.

The analysis was produced from client session/conversion data using Excel pivot tables, two‑sample z‑tests, and exported charts/reports for stakeholders.

Data is processed and aggregated in Excel (pivot tables + formulas). Country and page-variant aggregations compute conversion counts and rates, a two‑sample z‑test evaluates significance, plots (PNG) visualize totals and country breakdowns, and results are compiled into a PDF/Word report for stakeholder review.

• Large-sample, production-scale analysis: the dataset covers ~294,000 sessions and ~35k conversions, enabling robust aggregate comparisons (overall difference ≈ 0.17% favoring the old page).
• Rigorous statistical validation: a two-sample z-test (z ≈ -0.047, two‑tailed p ≈ 0.963) is implemented and documented, demonstrating the observed difference is not statistically significant.
• Market-level segmentation: explicit country breakdowns (US, UK, Canada) quantify per-market performance and reveal the US dominates conversions while the old page generally outperforms or ties the variant.
• Reproducible Excel workbook & visuals: the repo includes the analysis workbook and exported images (Final_countries_ab.xlsx, report_sheet.png, Z_test.png, country_wise.png, Across_pages.png) so findings can be inspected and re-run by analysts.
• Stakeholder-ready deliverables and recommendations: a polished project_report.pdf and a Word brief (CR in Pages.docx) summarize findings, recommend retaining the old page, and propose follow-ups (longer test, device segmentation, logistic regression, BI dashboard).

I analyzed a hotel reservations dataset to identify booking patterns, cancellation drivers, guest demographics, and revenue opportunities so stakeholders could target marketing and reduce lost revenue from no-shows and cancellations. The data-backed analysis covered ~234.99K guests across years with seasonal and hotel-type differences highlighted.

I led end-to-end exploratory analysis and reporting: clean the dataset, identify key trends and risk drivers (cancellations, lead time, repeat customers), build an interactive dashboard for stakeholders, and write a concise recommendations report. I used the cleaned dataset (hotel_booking_data_cleaned.csv) to run descriptive and segment analyses (by hotel type, room type, season, lead time, and agent/channel). I produced visualizations and an interactive Power BI dashboard (Hotel booking Dashboard.pbix) and authored a written report (hotel_booking_report.pdf / .docx).

Key steps: data validation and cleaning (produced the cleaned CSV), time-series and seasonality checks to find peak months, segmentation to compare urban vs resort behavior (lead time and cancellation rates), demographic breakdowns (adults, infants, nationalities), and meal-plan and agent-channel analyses to prioritize revenue levers. I synthesized findings into actionable recommendations (policy, loyalty, targeted marketing) and bundled visuals and narrative into the deliverables. Main challenges addressed: noisy/heterogeneous booking records and isolating channel- and season-specific patterns; I mitigated these by careful cleaning and segmented analysis.

• Deliverables: interactive Power BI dashboard (.pbix), cleaned dataset (CSV), and a written report (PDF/DOCX) that together make the analysis reusable for stakeholders.
• Key quantitative findings: total guests ≈ 234.99K; adult bookings ≈ 222K; repeat customers only 3.2% (with a high cancellation rate of 37%); urban lead time ≈ 109.7 days vs resort ≈ 92.7 days; peak booking months include August and December; Room Type A had the highest and seasonally varying demand.
• Business recommendations and impact: tighten or redesign cancellation policy and incentives to reduce no-shows; launch loyalty or targeted offers to lift the low 3.2% repeat rate; shift marketing spend to peak months (August/December) to capture demand surges; promote HB/FB meal plans and strengthen travel-agent relationships to increase revenue per booking. These steps aim to reduce cancellations, increase repeat business, and improve revenue capture from existing demand.

This project analyzes transactional and behavioral data from an e‑commerce platform to surface patterns in purchasing, product preference, returns, delivery performance, and satisfaction. Its purpose is to produce data‑backed recommendations that reduce return rates, boost retention, optimize discounts and delivery, and improve conversion to premium subscriptions. Originating as a Jupyter Notebook analysis (with HTML, PDF and PPTX exports), it uses a 10,000‑customer CSV dataset included in the repository for a practical business‑analytics study.

A Jupyter Notebook drives the workflow: Python (Pandas, NumPy) ingests the CSV, performs cleaning and feature encoding, computes z‑scores and correlations (Pearson), and runs grouped comparisons; Matplotlib/Seaborn and Plotly Express produce visualizations saved under images/, and outputs are exported as HTML, PDF, and PPTX for sharing.

• Data quality and preparation: validated a 10,000‑row, 16‑column dataset with no missing or duplicate rows; ambiguous columns were renamed and categorical satisfaction levels converted to numeric for analysis.
• Robust exploratory analysis: computed descriptive statistics, variance, z‑score outlier detection, and segment‑level summaries to characterize customer demographics and spend.
• Statistical tests & correlations: ran Pearson correlations (e.g., Time Spent vs Purchase Amount r ≈ 0.01) and grouped comparisons across payment methods, subscription status, device types, and regions to isolate effects.
• Actionable visualizations: produced targeted charts (category volumes, return rates, device distribution, delivery time by subscription, payment method satisfaction) and saved image assets for reporting.
• Deliverables & recommendations: packaged findings into notebook, HTML, PDF, and PPTX artifacts and distilled 10 concrete business recommendations (e.g., prioritize quality control for Electronics/Toys, redesign premium delivery promise, and replace flat discounts with tiered/bundled offers).

A client needed actionable insights from user tracking data to understand who their users are, how they access the site (device & session behavior), and where engagement drops off so they could prioritize UX and marketing improvements.

Solution: The project ingests a client-provided tracking dataset (track_dataset) and produces cleaned, analyzed, and visualized outputs using Google Sheets. It identifies usage patterns by country, device type, gender and visit duration, and highlights engagement outliers. Results are presented as interactive charts and a dashboard image plus a written report (PDF/DOCX) for stakeholders.

Visualizations and recommendations are the primary deliverables: a dashboard PNG, distribution charts (device, gender, country), session-duration analyses that flag potential bounce issues, and actionable guidance (cross-device optimization, retention-focused experiments). The README and attached report summarize findings and suggest next steps like predictive modeling and deeper statistical tests.

Tech Stack: Primary tool: Google Sheets (data cleaning, transformation, charting, dashboarding). File formats / artifacts: Markdown README, PNG dashboard images, PDF/DOCX report, source dataset (track_dataset referenced). No application server or database present in repo — analysis is sheet-driven and export-based ((Implied capabilities) Google Sheets charts/filters, pivot tables, and typical spreadsheet formulas for aggregation and visualization).

• Device usage distribution: Desktop accounts for ~36% of traffic, only slightly ahead of mobile and tablet. This close distribution implies marketing and UX must target multi-device parity; fixing cross-device layout/flows could reduce friction and improve conversions across all channels.
• Engagement variability: Session durations vary widely — some sessions >13 minutes while others are under 1 minute. This signals a mix of highly engaged users and probable bounce/low-engagement visits. Segmenting by device and country to identify drivers of short sessions can guide retention improvements and prioritize UI fixes.
• Geographic insights: Analysis covers users across four countries (README). Country-level trends in device preference and session length can inform region-specific campaigns and localized UX adjustments to increase relevance and engagement.
• Deliverables and communication: The repo contains a dashboard image and a formal report (PDF/DOCX) suitable for stakeholder presentation, improving decision speed. Visual, shareable deliverables reduce time-to-action compared to raw CSVs.
• Key learnings & next steps: The project revealed the value of quick spreadsheet-driven analysis for exploratory insights. To increase impact, add: (a) deeper statistical tests (A/B or cohort analysis), (b) predictive models for churn/retention, and (c) automated ETL and reproducible scripts (Python/R or Apps Script) to scale beyond manual Sheets processes.

The project analyzed ~20 years of car-sales data to surface business-critical trends and KPIs (sales, profit, orders, quantity, brand and demographic patterns) so stakeholders could understand historical performance, pandemic impact (2020 dip), and customer preferences to guide strategic decisions. I owned the end-to-end analytics deliverable: obtain and clean the dataset, run exploratory and trend analyses, produce visuals and an interactive Power BI dashboard, and produce a written report with actionable recommendations.

Action: Sourced a Kaggle-style dataset (Cars Mock Data.csv), cleaned and preprocessed it in a Jupyter notebook using pandas/numpy (missing-value handling, type fixes, deduping, and outlier checks). Aggregated metrics by year, brand, model, and customer segments (age, gender, color). Performed time-series and year-over-year comparisons to quantify pandemic effects and recent declines. Computed totals (sales, profit), orders vs. quantity, top brands/models, and demographic splits.

Built interactive visualizations and KPI pages in MS Power BI for stakeholder exploration. Compiled findings, charts, and recommendations into a PDF and Word report. Challenges addressed included noisy/mock data, inconsistent formats, and skewed age distribution — solved with conservative imputations, grouped aggregations, and clear filtering options in the dashboard. Reproducible code and artifacts were included for handoff.

• Result: Key numbers: Total sales ≈ $553.5M and profit ≈ $360M over the period; 10,000 orders representing ~49,000 car sales; leading brand: Ford; orders declined ~8.7% in the most recent year; clear sales dip in 2020 attributed to COVID-19.
• Customer insights: Buyers concentrated in the 60–70 age band; women showed ~7.4% higher interest than men; top colors and Ford Mustang as a top model were identified to guide product/marketing choices.
• Deliverables: Jupyter notebook, Cars Mock Data.csv, Power BI dashboard , and written reports all included in the repo for reproducibility and stakeholder review.
• Impact: Provided evidence-based recommendations to prioritize targeted marketing, diversify customer-acquisition strategies to younger cohorts, and prepare contingency responses for demand shocks (e.g., pandemic), enabling data-driven planning and executive reporting.

A descriptive analysis project built around a laptop-inventory dataset to uncover sales trends, stock imbalances, and demand signals. The goal was to clean and harmonize heterogeneous product/specification data, quantify model/brand performance, segment inventory by demand and quality, and produce actionable recommendations and visualizations to reduce stockouts and excess inventory while improving stocking decisions.

Cleaned and preprocessed the raw inventory dataset, engineered metrics that capture product performance and stock quality, ran segmentation and comparative analyses by brand/processor/specs, produced visualizations and an interactive dashboard, and delivered a concise report with operational recommendations for inventory optimization.

• Ingested the raw CSV and performed exploratory data analysis; standardized categorical fields, parsed spec strings, handled missing values, deduplicated records, and engineered a composite Performance Index alongside stock-quality indicators using Pandas and NumPy.
• Aggregated and visualized metrics by brand, processor, and display category using Matplotlib and Seaborn to surface outliers and slow movers in time-agnostic cross-sections.
• Applied KMeans clustering with StandardScaling to segment products by performance and stock characteristics, iterating on feature selection to overcome noisy specification strings and small-brand class imbalances.
• Packaged key visualizations into an interactive report/dashboard and exported a written summary with prioritized stock actions.
• Quantified and delivered business impact: improved data hygiene and consistency, identified high-performing models for reorder alongside slow movers for discounting, aligned stock to market demand, and provided decision support to reduce stockouts and holding costs.

Context & Goal: The project analyzed health conditions across diabetic, prediabetic, and non‑diabetic populations to find patterns and correlations between diabetes status and factors such as BMI, blood pressure, cholesterol, smoking, alcohol use, diet, age, and gender. The goal was to surface actionable insights for prevention, screening, product development, and targeted health campaigns.

My Role: Conduct an end‑to‑end analysis: clean and explore the population health dataset, quantify prevalence and demographic patterns, investigate risk‑factor correlations with diabetes status, produce visualizations and a written report that translates findings into healthcare and product/marketing recommendations.

• Ingested and preprocessed the large population dataset, handling missing values, inconsistent encodings, and outliers (age/BMI anomalies) so comparisons were valid.
• Performed exploratory data analysis and group comparisons by diabetes status (non‑diabetic, prediabetic, diabetic): summary statistics, cross‑tabs, age‑stratified BMI trends, and gender breakdowns.
• Assessed prevalence and comorbidity relationships (high blood pressure, high cholesterol) and behavioral factors (fruit consumption, smoking, alcohol) using descriptive statistics and visual inspection.
• Built interactive visualizations and dashboards to surface trends and enable filtering by age/gender/status, and compiled a narrative report with key charts and interpretation.
• Addressed challenges from class imbalance and scale by stratifying analyses, using proportional summaries, and focusing on effect sizes rather than only p‑values. Tools: data processing and stats libraries plus a BI/visualization tool to present findings and produce the final report.
• Quantitative Snapshot: Non‑diabetic: ~197,000 patients; Diabetic: ~34,000; Prediabetic: ~6,000. Gender among diabetics roughly balanced (~17.5K males, ~16.1K females). Notable BMI pattern: very high average BMI in the youngest cohort examined (~36.16) declining across older age bands (~28.43 in later cohort). Older age bands show higher rates of hypertension and high cholesterol; a large portion of non‑diabetics report regular fruit consumption (~63%).
• Qualitative Impact & Recommendations: Evidence supports targeted early screening and prevention programs focused on younger high‑BMI groups and older patients with elevated blood pressure/cholesterol. Product opportunities for weight‑management, personalized nutrition, and lifestyle interventions aimed at age and risk segments. Marketing and outreach should be gender‑inclusive given balanced prevalence; emphasize preventive services and dietary counseling.
• Deliverables: an interactive dashboard and a written report that stakeholders (clinicians, public‑health teams, product leads) can use to prioritize interventions and monitor trends.

This project focused on improving supply chain performance by optimising product manufacturing, transportation, supplier selection, and warehouse productivity. Using a small, structured dataset, the goal was to identify supply, product, and transportation patterns that influence restocking speed, cost trade-offs, quality issues, and inventory shortfalls, enabling stakeholders to prioritise suppliers, shipping modes, and quality control actions.

You were responsible for preparing and analysing the dataset in a Jupyter Notebook, cleaning and validating the data, performing supplier, product, and transportation analyses, and translating the findings into clear, actionable recommendations to improve restocking, transportation decisions, inventory levels, and product quality.

Action:
• Loaded the provided dataset into a Jupyter Notebook and imported the necessary data analysis libraries. Inspected columns and missing values, removed irrelevant columns, and imputed missing values using best-matching records to preserve data integrity. Detected and handled duplicate entries to avoid bias from repeated observations.
• Performed supplier-level analysis to compare lead times and shipping costs, identifying suppliers best suited for rapid restocking versus cost-effective bulk orders. Conducted product category analysis to compare prices, stock levels, production volumes, and defect rates across skincare, cosmetics, and haircare categories.
• Evaluated transportation modes based on speed, safety, cost, and defect rates, and examined geographic usage patterns to determine where high-value shipments benefit from faster transport. Synthesised the findings into operational insights and prioritised recommendations that balanced urgency, cost, and quality constraints within the limitations of the dataset.

Result:
• Prioritised suppliers by identifying one supplier best suited for urgent restocking due to the shortest lead time and another for cost-effective bulk orders. Recommended using air transport for high-value or time-sensitive shipments, while reserving rail and sea transport for bulk shipments to balance cost and delivery speed.
• Recommended increasing inventory levels for high-cost, high-demand skincare products by addressing manufacturing lead-time variability and improving demand forecasting. Identified haircare as having the highest defect rates and recommended strengthening quality control processes to reduce defects and improve product availability. Produced a concise set of actionable recommendations and a comprehensive written report summarising the analyses and their operational implications.

This Jupyter Notebook analyzes customer transaction data to uncover purchase patterns, segment customers, and extract association rules for cross-sell opportunities. It is designed to provide actionable insights and simple recommendation logic so business teams can prioritize offers and marketing based on observed buying behavior. Originating as an exploratory data-science notebook, it bundles data cleaning, feature engineering, unsupervised modeling, and visualizations for rapid hypothesis testing.

A single-notebook pipeline: ingest transactional CSVs, clean and aggregate into customer–item matrices, run EDA and feature engineering (recency/frequency/monetary, basket encoding), apply clustering (e.g., K-means/DBSCAN) and association-rule mining (Apriori/FP-growth), then visualize segments and rules using matplotlib/seaborn; implemented with pandas, scikit-learn, mlxtend, and notebook-native visual outputs for reproducibility.

• End-to-end, reproducible analysis in one notebook: data loading, preprocessing, modeling, and visualization are organized into clear, runnable sections that facilitate review and iteration.
• Robust preprocessing: handles missing values, categorical encoding, basket pivoting and RFM feature construction to produce clean inputs for both clustering and association-rule algorithms.
• Multi-method insight extraction: combines clustering for customer segmentation with association-rule mining to surface frequent itemsets and rule-based cross-sell suggestions.
• Interpretable outputs and visualizations: segment profiles, cluster centroids, support/confidence/lift tables, and annotated plots make findings accessible to non-technical stakeholders.
• Lightweight, extensible tooling: uses standard Python libraries (pandas, scikit-learn, mlxtend, matplotlib/seaborn) so models and rules can be exported, parameterized, or refactored into a modular pipeline or production service later.

The project analyzes a social-media ad campaign dataset to understand which factors drive conversions and revenue. The goal was to use cluster and trend analysis to reveal how age, gender, campaign timing, channel, and campaign type affect conversions and profitability, and to surface actionable recommendations for optimizing spend and strategy.

I conducted an end-to-end data analysis: ingesting the dataset, running SQL queries and clustering, measuring investment vs. revenue across channels and campaign types, calculating engagement and opt-out rates, and producing a written report with findings and recommendations.

I sourced a public marketing dataset, loaded it into a relational DB for exploration, and wrote SQL queries to profile spend and revenue by channel and campaign type. I applied cluster analysis to segment users by demographic and timing features, compared investment-to-revenue ratios across channels, identified campaign pairs with high profitability, and computed opt-out and engagement metrics.

Based on the analytical results I interpreted trade-offs (e.g., high investment vs. high return), validated patterns across segments (age, gender, timing), and distilled operational recommendations to prioritize and optimize campaigns. Challenges addressed included normalizing campaign and channel categories for fair comparison and isolating conversion-focused signals from awareness/retention noise.

• Identified PPC and Referral as top revenue generators, with PPC delivering the highest revenue per total investment.
• Found Conversion campaigns produce the highest revenue but demand the largest budgets; Awareness campaigns follow with substantial, though lower, spend.
• Highlighted Referral promotions as particularly effective for both Awareness and Conversion objectives.
• Noted a near-50% email opt-out rate, indicating a sizable drop in sustained engagement and a need to improve targeting and follow-up.
• Recommended prioritizing Conversion and Awareness campaigns, focusing optimization on PPC/Referral channels, and revising retention tactics to improve long-term ROI.

The project performs an end-to-end, SQL-driven analysis of the U.S. short-term rental market to turn listing-level data (price, room type, availability, host counts, neighborhood, reviews) into actionable market insights across major U.S. metros. The goal was to reveal pricing trends, host concentration, availability patterns, and neighborhood-level demand to inform host acquisition and pricing strategies.

I was responsible for sourcing and preparing the dataset, building a reproducible analysis pipeline, designing and running SQL queries to explore host/price/availability/review patterns, and converting the analytic findings into clear, business-facing recommendations.

Action:

• Sourced the U.S. Airbnb open dataset from a public marketplace and preprocessed it in a notebook environment to clean prices, normalize room types, handle missing values, and resolve availability/reporting anomalies.
• Loaded cleaned tables into MS SQL Server to enable performant, repeatable SQL-based analysis and to centralize aggregations.
• Performed exploratory analysis with focused queries and aggregations: host counts by city, average price by room type, availability vs. host count, top neighborhoods by review volume and price, detection of zero-availability listings, and identification of low-priced listings with high review value.
• Addressed data-quality challenges such as misreported availability, inconsistent price formats, and sparse review histories by filtering, standardizing fields, and flagging suspect records for further review.
• Interpreted statistical patterns into business narratives (e.g., which listing types and metros drive demand) and formulated practical recommendations for hosts and platform strategy.
• Produced reproducible SQL queries and notebook artifacts to support stakeholder review and future updates.

Result:

• Cities with the largest host bases (examples): New York City ~45,742; Los Angeles ~31,530; Hawaii ~22,434; San Diego ~12,401; Austin ~10,433.
• Average price by room type: Hotel room $281; Entire home/apt $270; Private room $106; Shared room $73.
• Availability vs. host count: Entire home/apt — avg availability 165 days, total hosts ~154,157; Private room — 141 days, ~65,870 hosts; Shared room — 186 days, ~4,029 hosts.
• Top neighborhoods by review and price include high-review, premium neighborhoods (e.g., Presidio avg reviews 249 / avg price $120; Huguenot 198 / $180), indicating reputation supports premium pricing.
• Key insights: Entire-home listings dominate host volume and engagement; shared rooms show low price, low reviews, and high availability (low demand); hotel rooms occupy a small premium niche (high price, low host count).
• Business impact / recommendations (concise): prioritize entire-home recruitment and marketing, reassess and repurpose underperforming shared-room inventory, investigate zero-availability listings for data integrity or inactive hosts, use high-review neighborhoods as pricing benchmarks, and focus acquisition/marketing spend on top metros while piloting incentives in mid-tier markets.