LPG Cylinder Failure Prediction using Bigquery and ML: Implementation Summary
1. Overview
This project predicts LPG cylinder failure using machine learning and visualizes findings in a dashboard. It leverages BigQuery ML for modeling, Python (Google Colab) for data analysis, and Looker Studio for visualization. The goal is to identify at-risk cylinders and understand failure causes.
2. Problem Statement
The project addresses a stakeholder request for early failure detection. Key questions:
Success: A highly accurate prediction model.
Failure: An inaccurate model.
Key Trends: Factors contributing to cylinder failure.
3. Tools & Data
Talend ETL : extracts raw data from Oracle cloud
BigQuery: Stores and queries failure data.
Google Colab: Handles data analysis and model building.
Looker Studio: Creates an interactive dashboard.
Data: Includes manufacturer, manufacture date, past testing history, and visual inspection results.
4. Workflow
Data Preparation: Import historical and current failure datasets into BigQuery, merge them for analysis.
Model Building (Google Colab):
Use PyCaret for AutoML classification.
Compare models (e.g., Random Forest) for best performance.
Predict failures and write results back to BigQuery.
Feature Importance: Identify key failure factors using feature importance analysis.
Dashboard Creation (Looker Studio):
Display key metrics (failure percentage, manufacturer trends).
Visualize feature importance and failure distribution by manufacturer.
5. Key Insights
Success is defined by model accuracy.
Important trends highlight key failure causes.
AutoML (PyCaret) simplifies model selection.
Dashboard answers stakeholder questions and enhances data-driven decision-making.
6. Potential Improvements
Refining feature engineering and model tuning.

Enhancing dashboard usability with advanced filters and visual customizations.