[TBD]Knowing about ML troubleshooting

Introduction

Laymen explanation

Technical explanation

Causes

• • Data issue

  • Model issue

  • Hyperparameters issue

Detection tools

Confusion matrix - Refer here

Data visualisation

Listing issues and tips

Overfitting issue

It can be detected by result visualisation. Refer here for detail

Convergence issue

Refer here

Model performance issue

If linear model is used for non-linear dataset, then performance will not be good. Refer here for detail

This article helps to understand impact of model bias on performance.

Machine learning diagnostics

• Check if dataset represents the population and clean (low bias, low noise)

• For accuracy issue, you can take samples of misclassified input and see what could be added as extra features which can make ML classify correctly

  • Cross validation approach helps in generalisation. Refer here for detail

  • Confusion matrix related graphs (ROC curve, Precision/recall curve) helps to identify accuracy issue. Refer here for detail

Reference

https://coursera.org/share/6ac91da8bbc5589c98166efd2934fb19

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/knowing-what-makes-ml-training-converge

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/role-of-regularisation-in-machine-learning

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/selecting-ml-right-algorithm-for-your-problem

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/role-of-confusion-matrix-in-machine-learning

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/understanding-bias-in-machine-learning

https://coursera.org/share/2b4aee2b8310333755914e3cd5ceef56

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/role-of-cross-validation-data-in-machine-learning

https://sites.google.com/site/jbsakabffoi12449ujkn/home/machine-intelligence/role-of-confusion-matrix-in-machine-learning

https://towardsdatascience.com/problems-in-machine-learning-models-check-your-data-first-f6c2c88c5ec2