Accurate Numerical Classification Model
Budget: $2 – $8 USD
I have a purely numerical dataset that needs to be turned into a high-performing classification model, with accuracy as the top priority. I will share the raw data along with a short data dictionary; it still contains a few missing values and stray outliers that must be handled before modelling.
The workflow I expect is straightforward:
• Data preparation: clean, impute, normalise or standardise where appropriate, and generate any useful engineered features.
• Exploratory analysis: quick visual and statistical checks so we both understand the main patterns and potential pitfalls.
• Model development: train and fine-tune one or more algorithms in Python—TensorFlow, scikit-learn, XGBoost or any other library you feel fits—aiming for the best validation accuracy without overfitting.
• Evaluation & hand-off: provide metrics (confusion matrix, ROC-AUC, precision/recall, etc.), the annotated Python notebook or .py script, and clear instructions for reproducing results on my machine.
Clean, well-commented code and a concise summary report are the two key deliverables; anything that boosts interpretability—feature importance plots, SHAP values—is a plus. If additional compute is required, let me know up front so we can coordinate. Let’s build a model that hits the accuracy target while remaining deployable and easy to maintain.
The workflow I expect is straightforward:
• Data preparation: clean, impute, normalise or standardise where appropriate, and generate any useful engineered features.
• Exploratory analysis: quick visual and statistical checks so we both understand the main patterns and potential pitfalls.
• Model development: train and fine-tune one or more algorithms in Python—TensorFlow, scikit-learn, XGBoost or any other library you feel fits—aiming for the best validation accuracy without overfitting.
• Evaluation & hand-off: provide metrics (confusion matrix, ROC-AUC, precision/recall, etc.), the annotated Python notebook or .py script, and clear instructions for reproducing results on my machine.
Clean, well-commented code and a concise summary report are the two key deliverables; anything that boosts interpretability—feature importance plots, SHAP values—is a plus. If additional compute is required, let me know up front so we can coordinate. Let’s build a model that hits the accuracy target while remaining deployable and easy to maintain.