Water Potability Classification and Quality Clustering

Data analysis experimentation using the dataset from Kadiwal, A. (2020)

Methodology

• Data Source Collection
• Data Cleaning and Pre-processing
  • Train Test Split of 80/20
  • Creating 3 sets, Origial, MinMax, and Standard scaled
• Modelling
  • Nearest Neighbors
  • Decision Tree
  • K-means Clustering
• Result analysis
  • Best classification model
    • Metrics: Accuracy, Precision, Recall, F1 Score
  • Highest Scoring cluster
    • Silhouette Score

Tech stack: Jupyter for the modelling and charting, Canva for the images

Results

Classification

Best performing model is the Nearest Neighbors (k=10) with the Standard Scaling. This achieved an accuracy of 69.65% and precision of 71.43%.

Clustering

Across the multiple features, only 2 features (solids vs turbidity) are used in the visualisation but all features were used in the model.

K-means clustering with K=2 using the Original dataset performed best with a silhouette score of 0.571.