🚀 Spaceship Titanic - ML Classification Project

This notebook is a solution for the "Spaceship Titanic" machine learning competition on Kaggle. The task is to predict whether a passenger was transported to an alternate dimension based on various characteristics.

---

📁 Dataset Overview

The dataset is structured like a typical tabular classification problem:

Main files:

• train.csv: Includes ~8,700 passengers with labeled target (Transported)
• test.csv: ~4,300 passengers without labels

Key features:

• Categorical:
  HomePlanet, CryoSleep, Destination, VIP, Cabin
• Numerical:
  Age, RoomService, FoodCourt, ShoppingMall, Spa, VRDeck
• Target:
  Transported (Boolean: True/False)

---

🧠 Project Workflow

1. Exploratory Data Analysis (EDA)

• Checked class balance: the dataset is quite balanced between transported and not transported.
• Visualized distribution of numerical features like Age and Spa spending.
• Plotted categorical features using countplot() and pie charts to spot trends.
• Observed that features like CryoSleep, Destination, and VIP had visible correlations with the target.

2. Data Cleaning & Preprocessing

• Dropped:
  • Name: mostly irrelevant for prediction.
  • Cabin: many missing values, complex to parse initially.
• Handled missing values:
  • For categoricals like HomePlanet, filled with "unknown".
  • For numerical spendings (RoomService, FoodCourt, etc.), filled with 0.
  • For Age, used median imputation.
• Converted categorical to numerical using LabelEncoder.

3. Feature Engineering

• Simplified data types (e.g., from float64 to float32).
• Created no new features, kept baseline model clean and interpretable.

4. Model Training & Evaluation

• Trained and evaluated the following classifiers:
  • Logistic Regression
  • Decision Tree
  • Random Forest
  • Extra Trees Classifier
  • LightGBM (LGBMClassifier)
  • XGBoost (XGBClassifier)
• Used train_test_split for validation.
• Evaluated using:
  • accuracy_score
  • classification_report
  • confusion_matrix
• Compared each model’s performance. Tree-based models (especially XGBoost and LGBM) outperformed others in both accuracy and generalization.

5. Hyperparameter Tuning

• Applied GridSearchCV to:
  • LGBMClassifier
  • RandomForestClassifier
• Tuned parameters like:
  • n_estimators
  • max_depth
  • learning_rate (for LGBM)

---

🧪 Results

• Best model: LightGBM with tuned hyperparameters.
• Achieved over 0.80 accuracy on validation set.
• Saved predictions into CSV for Kaggle submission.

---

📦 Dependencies

pandas
numpy
matplotlib
seaborn
scikit-learn
xgboost
lightgbm

---

🎯 Goal

To build a solid baseline ML pipeline and test the performance of various models on a real-world-style dataset. This project is useful for practicing:

• End-to-end ML workflows
• Data cleaning strategies
• Comparison of classic and modern classifiers
• Working with structured data from competitions

---