Comprehensive Python ML Model Build
Budget: ₹12,500 – ₹37,500 INR
I have a mixed dataset—numerical and categorical—whose end-goal is straightforward: accurate prediction. I need you to take the raw file from first inspection all the way to a validated model I can reuse, while clearly showing me every step that gets us there.
Here’s the flow I expect. You will:
• Clean the data, manage missing values, encode categories and scale where needed.
• Perform exploratory analysis, visualising key relationships so I can understand what drives the target.
• Engineer features that improve predictive power and explain why they help.
• Flag any columns that may be sensitive; I’m not yet sure whether anonymisation is required, so please advise and mask if necessary.
• Train several algorithms in Python (pandas, NumPy, scikit-learn; XGBoost or LightGBM welcome), compare them with sound validation, and select the best.
• Evaluate the chosen model on a hold-out set, reporting metrics, feature importance and any limitations.
Deliverables
• A fully runnable Jupyter notebook or .py script with tidy, commented code
• Clear markdown cells or a brief report that explains your methodology and findings
• A README with environment details (requirements.txt / conda YAML) so I can reproduce results on my machine
Acceptance is complete when the notebook runs end-to-end, the model beats a simple baseline, and the documentation lets a non-technical teammate grasp what was done and why.
Here’s the flow I expect. You will:
• Clean the data, manage missing values, encode categories and scale where needed.
• Perform exploratory analysis, visualising key relationships so I can understand what drives the target.
• Engineer features that improve predictive power and explain why they help.
• Flag any columns that may be sensitive; I’m not yet sure whether anonymisation is required, so please advise and mask if necessary.
• Train several algorithms in Python (pandas, NumPy, scikit-learn; XGBoost or LightGBM welcome), compare them with sound validation, and select the best.
• Evaluate the chosen model on a hold-out set, reporting metrics, feature importance and any limitations.
Deliverables
• A fully runnable Jupyter notebook or .py script with tidy, commented code
• Clear markdown cells or a brief report that explains your methodology and findings
• A README with environment details (requirements.txt / conda YAML) so I can reproduce results on my machine
Acceptance is complete when the notebook runs end-to-end, the model beats a simple baseline, and the documentation lets a non-technical teammate grasp what was done and why.