Chenfeng LI

Performed a machine learning model to predict total sales for each product and shop for a retail chain using 2.9 million daily historical sales data from 22,169 kinds of items and 59 shops using NLP and time series techniques, enhancing inventory optimization and driving data-driven decision-making.

• Imported and merged table into pandas DataFrames, removed duplicates and imputed missing values.
• Implemented TF-IDF vectorizer to extract features from item and shop names, creating text-based features.
• Engineered lagged features and trend-based features for time series analysis.
• Conducted Exploratory Data Analysis (EDA), including visualization of target distribution and time trends. Used multivariate heatmaps to analyze numerical and categorical pairings.
• Applied mean encoding on categorical features and matrix factorization for text features.
• Constructed and trained pipelines with the above transformations and Ridge, XGBoost, and LightGBM regressors.
• Performed feature selection using Recursive Feature Elimination with Cross-Validation (RFECV) and optimized hyperparameters using Bayesian optimization.
• Evaluated models through cross-validation, analyzing Root Mean Square Error (RMSE).
• Predicted future outcomes and compiled comprehensive reports.

Python, Scikit Learn, Machine Learning Pipeline, NLP, TF-IDF, mean encoding, matrix factorization, Ridge Regressor, LightGBM Regressor, XGBoost Regressor, feature selection, hyperparameter optimization.

Developed an Azure-based system with a data scientist partner for sentiment analysis on 16,000 post-sale reviews for an online clothing store on social media. Ingested data with Data Factory pipeline, cleaned data using SQL in Synapse Analytic, and implemented NLP techniques with a BERT-based model in Azure Databricks for sentiment analysis. Concluded a satisfactory rate of 96% and trend upward over time.

• Configured an Azure Data Factory pipeline to ingest data from on-premise database to Azure Data Lake Storage.
• Cleaned and filtered data using SQL scripts in Azure Synapse Analytics.
• Designed a notebook in Azure Databricks for data processing, analysis, and visualization.
• Implemented NLP models for text cleaning, lemmatization, stop-word removal, and perform sentiment intensity analysis with BERT based model from Hugging Face Transformers.
• Trained and evaluated a random forest classifier model to predict the sentiment of the apparel reviews.
• Visualized results by time and item and created comprehensive reports.
• Updated the dataset with a scheduled trigger in Azure Data Factory on a weekly basis.

Python, Database management, Azure Data Factory, Synapse Analytics, Databricks, NLP, Hugging Face, BERT.

Implemented a hybrid cloud architecture that connects an on-premises data center to an AWS VPC using a Site-to-Site VPN with configuring multiple AWS services. Solved the challenge of seamless, secure, and scalable integration between two environments improving operational efficiency and data protection.

• Set up VPCs in two regions with appropriate subnets, route tables, and internet gateways.
• Launched Amazon Linux 2 instances in both regions with secure groups allowing SSH and ICMP traffic.
• Established Site-to-Site VPN connection using virtual private gateway with Openswan configuration.
• Utilized Amazon S3 for critical data backup, enabling versioning and server-side encryption.
• Implemented IAM roles and policies for secure access management, ensuring least privilege principles.
• Monitored instance performance, network traffic and VPN connection using AWS CloudWatch.
• Configured Amazon SNS to send alerts for critical events.
• Automated infrastructure deployment with AWS CloudFormation for consistent, repeatable setups.

AWS Instances, security groups, VPC, Site-to-Site VPN, IAM, CloudWatch, SNS, CloudFormation.

Developed an advanced Power BI dashboard for an international sports corporation to analyze sales data, providing real-time insights into sales performance, discount analysis, and regional success. The dashboard was designed to facilitate data-driven decision-making by presenting key metrics in an interactive and user-friendly format.

• Created and filtered data using SQL on the on-premise dataset, leveraging a star schema data model with the Sales table at the center for optimized performance.
• Cleaned and transformed data using Power Query, ensuring data accuracy and consistency.
• Implemented advanced DAX formulas to create columns and measures for in-depth analysis, including update time display, discount calculations and fiscal year-specific insights.
• Designed a one-page interactive dashboard with key metrics, including total sales, customer counts, product sales, and discount analysis.
• Incorporated various visualizations for comprehensive sales insights, making the data easily interpretable.
• Enabled scheduled refresh to ensure real-time data updates, keeping the dashboard always up-to-date with the latest data.
• Published the dashboard to the Power BI service for broad access and distribution within the organization.
• Collaborated with stakeholders to understand business requirements and tailor the dashboard to meet their needs.
• Conducted user training sessions to ensure effective use and interpretation of the dashboard.

Microsoft SQL Server, Power BI services, Power Query, DAX.

Developed a comprehensive Power BI dashboard to analyze retail chain transactions, providing actionable insights into product sales, customer behavior, seasonal trends, and the effectiveness of promotions. The dashboard facilitated data-driven decision-making with detailed visualizations and interactive features.

• Utilized Power Query to clean and transform data, including splitting and unpivoting the Product column.
• Developer advanced DAX formulas to create calculated columns and measures for in-depth analysis.
• Created detailed Product Analysis and Customer Analysis pages featuring key metrics, slicers, and interactive　visuals such as line charts, treemaps, and ribbon charts.
• Ensured alignment and consistency across all dashboard pages for a cohesive user experience.
• Integrated product and customer analysis visuals into a combined Retail Analysis Page with interactive buttons for toggling between views using bookmarks.
• Designed and presented slides summarizing key insights and findings to stakeholders, facilitating informed decision-making.
• Conducted stakeholder meetings to gather requirements and incorporate feedback into the dashboard design.
• Provided training and support to users for effective utilization and interpretation of the dashboard.

Power BI services, Power Query, DAX, Bookmarks, Slides.