Overview / Usage

The project's primary objective is to utilize Intel OneAPI libraries to develop an optimal method for assessing water quality. This model has the potential for deployment across various domains, enabling classification to determine the safety of water for consumption.

Methodology / Approach

We approached the problem in the following manner:

• EDA- removed the features of less importance and transformed the data into the desired format
• Synthetic data generation- As the dataset was imbalanced we generated some synthetic data and added that to the original dataset to balance it
• Hyperparameter tuning- Many bayesian based strategies like hyperopt, tpot, optuna, etc. were used for getting the best hyperparameters.
• Training and inference- Two models were tested on the dataset:
  • pyDeepInsight- a CNN based model which converts tabular data into image and then uses pretrained models like efficientnet, etc., to do the classification. The score of this models on the test data was not satisfactory enough.
  • Stacking ensemble- In this method we stacked the following machine learning models to use ensemble learning:
    • Cat Boost
    • TabNet
    • LGBM
    • XGB
    • AdaBoost
    • Extra Trees
    • Gradient Boosting
    • Decision Tree
    • Random Forest

Technologies Used

Programming Language: Python

• Libraries:
  • scikit-learn-intelex
  • modin
  • intel-pytorch
  • tpot
  • optuna
  • ctgan
  • hyperopt
  • pyDeepInsight

Repository

https://github.com/02shanks/Water-Quality_Prediction_Intel-oneAPI