Fuel Cell Performance Prediction

This repository contains code to preprocess and build predictive models for a fuel cell performance dataset. The primary goal is to:

1. Select the correct target column based on the "roll number" based selection function.
2. Clean, preprocess, and transform the dataset.
3. Split the dataset into training and test sets (70/30).
4. Run multiple regression models using PyCaret to find the best-performing model.
5. Evaluate the selected model on the test set using standard regression metrics.

Dataset Description

1. Features (F1, F2, ..., F15): Various numerical input features that influence fuel cell performance.
2. Targets (Target1, Target2, Target3, Target4, Target5): Multiple potential target columns. The correct target is selected based on the last digit of the roll number: Roll number ends in 0 or 5 → Target1 Roll number ends in 1 or 6 → Target2 Roll number ends in 2 or 7 → Target3 Roll number ends in 3 or 8 → Target4 Roll number ends in 4 or 9 → Target5 After determining which target applies to a given row, all other target columns are dropped, leaving a single Target column for modeling.

Prerequisites

1. Python 3.x
2. Jupyter Notebook or Google Colab environment recommended
3. PyCaret for model training and evaluation
4. Standard Python libraries: pandas, numpy, matplotlib, seaborn, scikit-learn

Steps Performed by the Code

1. Data Loading: -Reads the dataset from the specified CSV file.

2. Preprocessing: -Fills missing numeric values with the mean.

-Selects only numeric columns for modeling.

-If any columns are objects and can be numerically coerced, they are converted.

-Optional: Plots histograms and a correlation heatmap for exploratory data analysis.

3. Target Selection Logic: -Assigns a pseudo-roll number based on the DataFrame index.

-Uses the last digit of this pseudo-roll number to decide which TargetN column to use.

-Creates a single Target column and drops the other target columns.

4. Modeling with PyCaret: -Sets up a regression environment with PyCaret.

-Compares multiple regression models (e.g., Linear Regression, Random Forest, XGBoost, etc.) based on their R² scores.

-Selects the top-performing models and displays their performance metrics.

5. 70/30 Train-Test Split: -Splits the dataset into 70% training and 30% testing using train_test_split.

-Finalizes the best model from the comparison step.

6. Evaluation on the Test Set: -Uses the finalized model to make predictions on the test data.

-Prints out performance metrics like R², RMSE, and MAE to evaluate model performance.

Results and Interpretation

1. Best Model: Bayesian Ridge Regression achieved the best performance with:

   R²: 0.251 – Explains 25% of the variance in the target variable.

   RMSE: 0.0582 and MAE: 0.0487 – Indicating close predictions to actual values.

2. Model Comparison: Bayesian Ridge outperformed other models, while tree-based and ensemble methods showed suboptimal results due to possible non-linearity or feature issues.

3. Feature Correlation: High correlations among features suggest redundancy, and the target variable showed weak correlations, contributing to low R².