MLP Assignments

This repository contains Machine Learning and Data Analysis assignments with comprehensive dataset analysis and preprocessing.

Project Structure

Week 1: Dataset Analysis & Querying

• Dataset: week1/Week1_GA_dataset.csv (10,000 × 12 real estate data)
• Notebooks:
  • Week1.1_Dataset_Analysis.ipynb: Data cleaning, missing value analysis, row/column queries
  • Week1.2_Dataset_Analysis.ipynb: Indexing, slicing, conditional filtering operations

Key Results Week 1.1:

• Unknown values: 5,548 total (1,823 literal "?" + 3,725 original NaN)
• Missing value analysis and data cleaning operations
• Row filtering based on missing value thresholds

Key Results Week 1.2:

• Even/odd row and column extraction
• Year-based filtering and conditional queries
• August property counts and locality-based price analysis

Week 2: Machine Learning Preprocessing

• Dataset: week2/GA_2_dataset.csv (10,000 × 13 gaming engagement data)
• Notebook: Week2_Dataset_Analysis.ipynb

Preprocessing Pipeline:

1. Data Type Analysis: Identified object columns (Gender, Location, GameGenre, GameDifficulty, EngagementLevel)
2. Missing Value Handling: 3,337 total null values across Age, Location, InGamePurchases, GameDifficulty
3. Imputation Strategy:
   • Age: Mean imputation from training data
   • Location: NaN → "Other"
   • GameDifficulty: Mode imputation
   • InGamePurchases: NaN → 0
4. Feature Engineering:
   • Ordinal encoding: GameDifficulty (Easy=0, Medium=1, Hard=2)
   • One-hot encoding: Gender, Location, GameGenre (drop_first=True)
   • StandardScaler: All numerical features
5. Train-Test Split: 80-20 split, random_state=42

Final Answers:

• Q1: Object columns: Gender, Location, GameGenre (from specified list)
• Q2: Males from Europe with purchases: 299
• Q3: Under-18 players with >10h playtime: 453
• Q4: Total null values: 3,337
• Q5: Least frequent target class: High (1,996 samples)
• Q6: Sum of transformed Age (test): 16.50 (standardized) | 63,585.24 (imputed raw)
• Q7: Sum of first 5 transformed rows: -7.17 (full scaling) | 6.84 (numeric-only scaling)

Technical Implementation

Dependencies: pandas, numpy, scikit-learn
Preprocessing Approach: Complete-case statistics → Imputation → Encoding → Scaling
Validation: Multiple verification cells ensure reproducibility and cross-check different interpretations

Methodology Notes

The notebooks include comprehensive verification sections that explore different interpretations of "transformed" data:

• Raw vs. standardized feature values
• Numeric-only vs. full-feature scaling approaches
• Complete-case vs. all-data statistics for imputation

All results are reproducible with the provided random seeds and preprocessing steps.