ralucan / prodco-ds

DS project part of WIDS 2020

PRODCO Defect Detection Project

This project was initially conducted as part of Accenture's 2020 Women in Data Science Accelerator program (see folder Women_in_Data_Science_programme_(February_to_April_2020) for the original files.

After more than 3 years, I am looking to explore new modelling solutions for this root cause analysis.

Overview

This project aims to perform a root cause analysis and develop a defect detection model for PRODCO, a multinational manufacturing company that produces high tech products for different industries. PRODCO has been experiencing high defect rates in their new integrated production line, which has caused massive additional costs and customer dissatisfaction. The project has two equally important parts: data analysis and analytical modelling, and business value and recommendations.

Data

The data for this project consists of production data for the last 3 months with 75,000 records. The data contains information about each item that has gone through the manufacturing process, such as SKU ID, path (position & environment in zone) through the production line, time elapsed in and between zones, and pass/defect type recorded by quality control. The data dictionary and maintenance costs are also provided by PRODCO.

Methodology

The methodology for this project follows these steps:

• Data exploration and preprocessing: explore the data to understand its structure, distribution, and quality. Perform any necessary data cleaning, transformation, and feature engineering steps to prepare the data for modelling.
• Data analysis and visualization: perform descriptive and inferential statistics on the data to gain insights into the production process and the defect types. Use appropriate visualization techniques to present the findings and highlight any patterns or anomalies.
• Modelling and evaluation: build and compare different analytical models to detect defects and identify their root causes. Use appropriate evaluation metrics and techniques to assess the performance and accuracy of the models.
• Interpretation and explanation: interpret and explain the results of the models in a clear and concise way. Use techniques such as feature importance, partial dependence plots, or SHAP values to show how the models make predictions and what factors influence them.
• Business value and recommendations: quantify the potential business value of implementing the models in terms of cost savings. To provide actionable recommendations for PRODCO to improve their production process and reduce defect rates.

Results

The results of this project are presented in a report that contains the following sections:

• Executive summary: a brief overview of the project objectives, methodology, results, and recommendations.
• Introduction: a detailed description of the project background, problem statement, scope, and objectives.
• Data analysis: a comprehensive presentation of the data exploration, analysis, and visualization steps and findings.
• Modelling: a thorough explanation of the modelling approach, techniques, results, and evaluation.
• Interpretation: a clear interpretation of the model outputs and their implications for defect detection and root cause analysis.
• Business value: a quantification of the business value of implementing the models and their impact on PRODCO's performance and profitability.
• Recommendations: a list of practical and feasible recommendations for PRODCO to improve their production process and reduce defect rates.
• Conclusion: a summary of the main findings, limitations, and future work of the project.

Code

The code for this project is written in Python using Jupyter Notebook. The code is organized into separate notebooks for each step of the methodology. The code is well-documented and follows best practices for readability and reproducibility. The code can be found in the code folder of this repository.