DEV Community

Cover image for How to Approach a Machine Learning Problem ?
SOMYA RAWAT
SOMYA RAWAT

Posted on

How to Approach a Machine Learning Problem ?

Step-1 : Frame the Problem

As a first step, you need to arculate your problem by
identifying the type which depends on your business
problem.

Type can be anything like Binary classification,
Unidimensional regression, Multi-class single-label
classification, Multi-class multi-label classification,
Multidimensional regression, Clustering (unsupervised),
Other (translation, parsing, bounding box id, etc.)

Step-2 : Get the Data

  • The next step is to get the data and store it the right
    format according to your problem statement.

  • Analyse your data to check whether you have enough
    data or not and also check the quality of the data.

  • The quality of the data fundamentally determines if
    you will be able to solve the problem at all or not.

Step-3 : Data Pre-Processing

  • After having the data next step is to analyse it and
    extract insights to make business decisions.

  • Also, apply basic data preprocessing operations to
    bring the data in a go to go format.

  • Choose the right library.

Step-4 : Evaluation Metric

  • The most important step is to know how to evaluate
    our results.

  • We need to choose the right evaluation metric
    according to the problem we are going to solve.

  • For example - if we have an imbalance dataset then
    we usually choose the ROC-AUC metric.

Step-5 : Split the data

  • In any machine learning problem, we split the data
    into multiple sets like training, validation and test.

  • Stratified splitting is mostly used for classification
    problems and K-Fold for regression problems.

  • The most important thing to note is whatever
    operations you apply on the train set must be
    applied to validation and test set.

Step-6 : Apply ML Algorithm

  • And finally, we will apply ML models to the data. We
    can't say which models work best it's just hit and
    trial.

  • Applying multiple algorithms do hyperparameter
    tuning, evaluate the results and choose the best
    model which gives satisfying results.

  • Benchmark your solution based on your selected
    evaluation metric.

Top comments (0)