Welcome to the Multi-Task Tool Prediction competition! Your goal is to solve three inter-dependent prediction tasks simultaneously using a synthetic dataset of 140,000 hardware tools.
The three tasks are:
* Clustering: Predict Island (8 store sections).
* Classification: Predict ToolType (25+ tool types).
* Regression: Predict Price (continuous value in rubles).
These tasks are stacked, meaning better clustering leads to better classification, and better classification leads to better price prediction. The best solutions will find ways to leverage all three tasks together.
The dataset contains 140,000 synthetic records of hardware tools with 21 features. The data is procedurally generated with realistic distributions, correlations, and dependencies.
The features are categorized as follows: * Basic: Company, Color, Country, Material * Physical: Weight, Dimensions_X, Dimensions_Y, Dimensions_Z, Power, Torque, Speed_Count, Reverse * Battery: Battery_Capacity, Battery_Voltage, Battery_Type * Subjective: Ergonomics, Durability, Reliability * Store: Store_X, Store_Y
Submissions are evaluated using a combined score that measures performance on all three tasks:
Score = 0.3 × (1 - Island_Accuracy) + 0.3 × (1 - ToolType_Accuracy) + 0.4 × Normalized_RMSE_Price
Where:
* Island_Accuracy: Accuracy of Island predictions (0–1)
* ToolType_Accuracy: Accuracy of ToolType predictions (0–1)
* Normalized_RMSE_Price: RMSE / (max_price - min_price), scaling the error to a 0–1 range
Lower score is better. The normalized RMSE ensures all components are on a comparable scale.
Your submission must be a CSV file with exactly 4 columns:
* Id: Tool ID (from test.csv)
* Island: Predicted island (must match one of 8 categories)
* ToolType: Predicted tool type (must match one of 25+ categories)
* Price: Predicted price (any positive float)
All three predictions must be submitted together in a single CSV file.