Diamond Dynamics Streamlit App

Project Takeaways:

• Data Cleaning & Preprocessing
• EDA & Data Visualization
• Feature Engineering
• Outlier & Skewness Handling
• Regression using ML algorithms, including ANN
• Dimensionality Reduction with PCA
• Clustering & Cluster Labeling
• Streamlit UI Design

Work Flow

• Data Insertion and Preparation data_exploration.ipynb
  • Data Preprocessing
  • Feature Engineering
    • Handle x,y,z values which were 0
    • Derived new columns such as Volume,Dimension Ratio.etc
  • Encoding
    • Created Ordinal Encoding for columns Cut,Clarity and Color
    • Created Pickle file to dump the encoders
• Exploratory Data Analysis EDA.ipynb
• Regression Model Reg_model_training.ipynb
  • Created 9 different Regression Algorithms and Chose the best Model for Deployment to predict Diamond Prices
    • Linear Regression
    • KNN Regressor
    • Support Vector Regressor
    • Decision Tree Regressor
    • Random Forest Regressor
    • Ada Boost Regressor
    • Gradient Boost Regressor
    • Xg Boost Regressor
    • ANN
      • Mean Squared Error Loss Function
      • Huber Loss Function

- Saved the Best Performing Model in a Pickle File

• Clustering Model cluster_model.ipynb
  • Used PCA for dimensionality reduction to 2 components
  • Created 5 different Clustering Model and chose the best model for deployment to cluster Diamond, then analyze average price, carat, and cut distribution per cluster
    • K-Means Clustering
      • Performed Elbow Method to get the K-value
    • Mini Batch K means Clustering
    • Hierarchical Clustering
    • Density Based Clustering
    • Gaussian Mixture
  • Saved The model with best Silhoutte Score in a Pickle File

• S3 Bucket

  • Saved all the encoders and models in the S3 Bucket.
  

• Streamlit UI (Application) streamlit.py

  • Retrieved all the files from S3 bucket for final predictions.
  • Created Input Features with sliders and number input boxes
  • Predicted the input values