Problem framing is a critical thinking process before starting an ML project (and any other research projects). Basically, researchers spend about 70-80% of entire research time to define and fine tune the direction of their researches. Imagine like preparing a blueprint of a building that requires information and modification until the foundation is solid. The design of the building can be expanded and customized based on new inputs later.
But the main point of doing problem framing in ML is to decide whether you even need ML in your project.
Follow the step-by-step guide for conducting problem framing.
- Context
Provide context from the real-world situation.
Explain what is actually happening and why this problem matter. - Question
Convert context in a simple single answerable question.
e.g. Can fabric appearance estimate exposure time to local microclimate?
ML Justification Check – Determine whether ML is necessary or if a simpler method can solve the problem.
- Define input-output mapping.
Specify what data goes in (input) and what the model should predict (output).
X(Features): Image of larval head capsules.
Y(Label): Larval age/instar - Select ML task type
Determine the appropriate ML (classification, regression, clustering). - Identify data representation
Decide how data is structured and encoded.
e.g. image of larval cephalopharyngeal skeleton labelled according to species. - Define success metrics
Establish how model performance will be measured and its meaning in biology.
Classification > accuracy, F1
Regression > RMSE, MAE
Clustering > Silhouette
For undergraduate project proposal, indicate success criteria based on input and output e.g. The model successfully classify cephalopharyngeal skeleton according to species. - Identify assumptions, limitations and biases
Explicitly state any underlying assumptions, limitations and biases in the study.
E.g. Sampling bias, unknown environmental data. - Check feasibility
Assess if the project solvable within time and provided resources. - Iterate
Based on 7 and 8, revisit 3, 4, 5 and 6.
Practice Problem Framing using this template.