koushik-l / industrial-copper-modeling

Industrial Copper Modeling

Project Domain - Manufacturing

Problem Statement:

In the copper industry, handling sales and pricing data posed challenges due to skewness and noise. Manual efforts were time-consuming and less accurate. Machine learning regression models were proposed to enhance accuracy by employing techniques like data normalization and outlier detection. Additionally, a lead classification model aimed to assess the likelihood of leads becoming customers, using WON for success and LOST for failure.

Problem Approach:

1. Data Understanding:

   • Identified variable types (continuous, categorical).

2. Data Preprocessing:

   • Handled missing values (mean/median/mode).
   • Treated outliers (IQR method).
   • Addressed skewness with transformations (log transformation).
   • Encoded categorical variables (label).

3. EDA:

   • Visualized outliers and skewness using Seaborn (boxplot, distplot, violinplot).

4. Feature Engineering:

   • Understood the feature and its importance.
   • Dropped highly correlated columns using a heatmap and coorelation factor.

5. Model Building and Evaluation:

   • Split dataset (training, testing).
   • Trained and evaluated by using ML algorithms (DecissionTree, RandomForest, XGBoost).
   • Optimized model hyperparameters (cross-validation, grid search).
   • Assessed performance (accuracy, precision, recall, F1, AUC).

6. Model Pickling

   • Used pickle to dump/load ML models.

7. Model GUI (Streamlit):

   • Created an interactive streamlit application for Regression/Classification model.
   • Values excluding 'Selling_Price' for regression, 'Status' for classification were taken as input.
   • Predicted output data by using ML models and then displayed results on streamlit.

Tools Used

• Python scripting
• Data Preprocessing
• EDA
• Streamlit