All projects
Machine Learning
Marketing Response Analysis
Result
Tuning improved responder recall from 28% to 47%, trading off precision - a deliberate shift toward catching more likely responders
Stack
PythonDecision TreeGrid Search CVpandasscikit-learn
01 - The problem
Campaign spend was being allocated on intuition, with no reliable read on which market signals actually preceded conversion.
02 - Approach
Built a Decision Tree classifier over historical campaign and market-trend data, then tuned it with Grid Search CV across depth, splitter and criterion parameters to improve responder recall.
03 - Outcome
Identified market-trend signals as the strongest predictors of campaign response, ahead of raw spend level and channel mix, giving the business an evidence base for reallocating budget instead of repeating last quarter's plan.