Economic Day Status Prediction

🏆 Achievement

Rank 1st in Public Score with 0.447021 RMSE

Prerequisites

Software Requirements

• Python 3.7+
• Jupyter Notebook or JupyterLab
• Git (for version control)

Required Libraries

pip install pandas>=1.3.0
pip install numpy>=1.21.0
pip install scikit-learn>=1.0.0
pip install matplotlib>=3.3.0
pip install seaborn>=0.11.0
pip install hdbscan>=0.8.27

Alternative Installation

Install all dependencies at once:

pip install -r requirements.txt

Hardware Recommendations

• RAM: Minimum 8GB, Recommended 16GB+ (for handling large datasets with missing values)
• CPU: Multi-core processor recommended (Random Forest utilizes parallel processing)
• Storage: At least 2GB free space for datasets and model outputs

Knowledge Prerequisites

• Python Programming: Intermediate level
• Machine Learning: Understanding of classification algorithms, feature engineering
• Data Science: Experience with pandas, scikit-learn
• Statistics: Basic understanding of missing data patterns, cross-validation
• Domain Knowledge: Familiarity with economic indicators (helpful but not required)

Dataset Requirements

• Access to the competition datasets (train.csv, test.csv)
• Preprocessed data from 00_DATA PREPARATION.ipynb
• Sufficient disk space for intermediate files and model artifacts

Overview

This project tackles a multi-class classification challenge to predict daily economic status across diverse global regions. The unique challenge involves handling datasets with 75% missing values (MCAR - Missing Completely At Random) from 10 different regions spanning Europe, Asia, America, and Africa.

📊 Problem Statement

• Task: Multi-class classification for economic day status prediction
• Target: economic_day_status with classes: Low, Medium, High
• Challenge: 75% missing data across all features
• Regions: 10 diverse datasets representing global economic conditions
• Evaluation Metric: F1-macro score

🗂️ Project Structure

├── Data/
│   ├── train.csv                    # Combined and imputed training data
│   └── test.csv                     # Combined and imputed test data
├── 00_DATA PREPARATION.ipynb        # Data combination and imputation
├── 01_EXPLORATORY DATA ANALYSIS.ipynb  # Data exploration and insights
├── 02_FEATURE ENGINEERING AND MODELING.ipynb  # This notebook
├── 0.447021-submission.csv          # Final predictions
└── README.md

🔧 Methodology

Data Preprocessing

• Missing Data Handling: Advanced imputation techniques for 75% MCAR data
• Feature Engineering: Strategic creation of economic indicators
• Encoding: One-hot encoding for categorical variables
• Data Integration: Combining 10 regional datasets

Feature Engineering Highlights

1. Temporal Features

# Lagged features for trend analysis
inflation_rate_lag_1
unemployment_rate_lag_1
exchange_rate_lag_1
business_confidence_index_lag_1

2. Rolling Statistics

# Moving averages for smoothing
inflation_rate_ma_7    # 7-day moving average
unemployment_rate_ma_7
exchange_rate_std_7    # Volatility measure

3. Economic Ratios

# Key economic indicators
exports_imports_ratio = exports_usd / imports_usd
debt_gdp_x_interest_rate = debt_gdp_ratio * interest_rate

4. Rate of Change Features

# Momentum indicators
inflation_rate_pct_change
exchange_rate_daily_change
biz_confidence_change

5. Threshold-based Binary Features

# Economic condition flags
high_inflation = (inflation_rate > 0.05)
high_unemployment = (unemployment_rate > 0.06)

6. Categorical Encoding

• Trade balance status: Deficit(-1), Neutral(0), Surplus(1)
• Political stability: Unstable(-1), Moderate(0), Stable(1)
• Income group: Low(-1), Lower-Middle(0), Upper-Middle(1)

Model Architecture

Random Forest Classifier with optimized hyperparameters:

RandomForestClassifier(
    n_estimators=805,
    max_depth=17,
    min_samples_split=14,
    min_samples_leaf=4,
    bootstrap=False,
    class_weight='balanced',
    random_state=42
)

Key Features Used

• Economic Indicators: GDP per capita, inflation rate, unemployment rate
• Trade Metrics: Exports, imports, trade balance status
• Financial Data: Interest rates, exchange rates, debt ratios
• Social Indicators: Education index, healthcare index, poverty rate
• Infrastructure: Internet penetration, urbanization percentage
• Governance: Political stability, governance quality

📈 Results

• Test F1-macro Score: 0.4359
• Public Leaderboard: Rank 1st with 0.447021 F1_Macro
• Model Performance: Robust handling of imbalanced classes with class_weight='balanced'

🛠️ Technical Implementation

Libraries Used

import pandas as pd
import numpy as np
import sklearn
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split, RandomizedSearchCV
from sklearn.metrics import f1_score
from sklearn.feature_selection import mutual_info_classif

Key Functions

• detailed_feature_engineering(): Comprehensive feature creation
• make_mi_scores(): Mutual information analysis
• Model training with cross-validation
• Automated submission file generation

🎯 Success Factors

1. Advanced Feature Engineering: Created 20+ new features from existing data
2. Missing Data Strategy: Effective handling of 75% missing values
3. Regional Diversity: Successfully integrated data from 10 different regions
4. Balanced Classification: Addressed class imbalance with proper weighting
5. Hyperparameter Optimization: Fine-tuned model parameters for optimal performance

📋 Usage

1. Data Preparation: Run 00_DATA PREPARATION.ipynb first
2. Exploration: Analyze data with 01_EXPLORATORY DATA ANALYSIS.ipynb
3. Modeling: Execute 02_FEATURE ENGINEERING AND MODELING.ipynb
4. Submission: Generated 0.447021-submission.csv for competition

🔍 Feature Importance

The model identified key economic indicators through mutual information analysis, with features like:

• Economic ratios and trade balances
• Temporal patterns in economic metrics
• Regional and governance factors
• Infrastructure and social indicators

🏅 Competition Highlights

• Successfully handled one of the most challenging aspects of real-world data: extensive missing values
• Achieved top performance through strategic feature engineering rather than complex ensemble methods
• Demonstrated effective cross-regional economic modeling
• Robust performance across diverse economic conditions globally

---

This project showcases advanced data science techniques for handling missing data and economic forecasting, achieving top-tier competition results through strategic feature engineering and robust modeling approaches.