Construction Cost Prediction with Machine Learning

Overview

Based on DDC methodology (Chapter 4.5), this skill enables predicting construction project costs using historical data and machine learning algorithms. The approach transforms traditional expert-based estimation into data-driven prediction.

Book Reference: "Будущее: прогнозы и машинное обучение" / "Future: Predictions and Machine Learning"

"Предсказания и прогнозы на основе исторических данных позволяют компаниям принимать более точные решения о стоимости и сроках проектов." — DDC Book, Chapter 4.5

Core Concepts

Historical Data → Feature Engineering → ML Model → Cost Prediction
    │                    │                │              │
    ▼                    ▼                ▼              ▼
Past projects      Prepare data      Train model    New project
with costs         for ML            on history     cost forecast

Quick Start

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_absolute_error, r2_score

# Load historical project data
df = pd.read_csv("historical_projects.csv")

# Features and target
X = df[['area_m2', 'floors', 'complexity_score']]
y = df['total_cost']

# Split data
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)

# Train model
model = LinearRegression()
model.fit(X_train, y_train)

# Predict
predictions = model.predict(X_test)
print(f"R² Score: {r2_score(y_test, predictions):.2f}")
print(f"MAE: ${mean_absolute_error(y_test, predictions):,.0f}")

# Predict new project
new_project = [[5000, 10, 3]]  # area, floors, complexity
cost = model.predict(new_project)
print(f"Predicted cost: ${cost[0]:,.0f}")

Data Preparation

Prepare Historical Dataset

import pandas as pd
import numpy as np

def prepare_cost_dataset(df):
    """Prepare historical project data for ML"""
    # Select relevant features
    features = [
        'area_m2',
        'floors',
        'building_type',
        'location',
        'year_completed',
        'complexity_score',
        'material_quality',
        'total_cost'
    ]

    df = df[features].copy()

    # Handle missing values
    df = df.dropna(subset=['total_cost'])
    df['complexity_score'] = df['complexity_score'].fillna(df['complexity_score'].median())

    # Encode categorical variables
    df = pd.get_dummies(df, columns=['building_type', 'location'])

    # Calculate derived features
    df['cost_per_m2'] = df['total_cost'] / df['area_m2']
    df['cost_per_floor'] = df['total_cost'] / df['floors']

    # Adjust for inflation (to current year prices)
    current_year = 2024
    inflation_rate = 0.03  # 3% annual
    df['years_ago'] = current_year - df['year_completed']
    df['adjusted_cost'] = df['total_cost'] * (1 + inflation_rate) ** df['years_ago']

    return df

# Usage
df = pd.read_csv("projects_history.csv")
df_prepared = prepare_cost_dataset(df)

Feature Engineering

def engineer_features(df):
    """Create additional features for better predictions"""
    # Interaction features
    df['area_x_floors'] = df['area_m2'] * df['floors']
    df['area_x_complexity'] = df['area_m2'] * df['complexity_score']

    # Polynomial features
    df['area_squared'] = df['area_m2'] ** 2

    # Log transforms (for skewed features)
    df['log_area'] = np.log1p(df['area_m2'])

    # Binned features
    df['size_category'] = pd.cut(
        df['area_m2'],
        bins=[0, 1000, 5000, 10000, float('inf')],
        labels=['small', 'medium', 'large', 'xlarge']
    )

    return df

Machine Learning Models

Linear Regression

from sklearn.linear_model import LinearRegression
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline

def train_linear_model(X_train, y_train):
    """Train Linear Regression model with scaling"""
    pipeline = Pipeline([
        ('scaler', StandardScaler()),
        ('regressor', LinearRegression())
    ])

    pipeline.fit(X_train, y_train)

    # Feature importance (coefficients)
    coefficients = pd.DataFrame({
        'feature': X_train.columns,
        'coefficient': pipeline.named_steps['regressor'].coef_
    }).sort_values('coefficient', key=abs, ascending=False)

    return pipeline, coefficients

# Usage
model, importance = train_linear_model(X_train, y_train)
print("Feature Importance:")
print(importance)

K-Nearest Neighbors (KNN)

from sklearn.neighbors import KNeighborsRegressor
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import GridSearchCV

def train_knn_model(X_train, y_train):
    """Train KNN model with optimal k"""
    # Scale features
    scaler = StandardScaler()
    X_scaled = scaler.fit_transform(X_train)

    # Find optimal k using cross-validation
    param_grid = {'n_neighbors': range(3, 20)}
    knn = KNeighborsRegressor()
    grid_search = GridSearchCV(knn, param_grid, cv=5, scoring='neg_mean_absolute_error')
    grid_search.fit(X_scaled, y_train)

    print(f"Best k: {grid_search.best_params_['n_neighbors']}")
    print(f"Best MAE: ${-grid_search.best_score_:,.0f}")

    return grid_search.best_estimator_, scaler

# Usage
knn_model, scaler = train_knn_model(X_train, y_train)

Random Forest

from sklearn.ensemble import RandomForestRegressor

def train_random_forest(X_train, y_train):
    """Train Random Forest model"""
    rf = RandomForestRegressor(
        n_estimators=100,
        max_depth=10,
        min_samples_split=5,
        random_state=42
    )

    rf.fit(X_train, y_train)

    # Feature importance
    importance = pd.DataFrame({
        'feature': X_train.columns,
        'importance': rf.feature_importances_
    }).sort_values('importance', ascending=False)

    return rf, importance

# Usage
rf_model, importance = train_random_forest(X_train, y_train)
print("Feature Importance:")
print(importance.head(10))

Gradient Boosting

from sklearn.ensemble import GradientBoostingRegressor

def train_gradient_boosting(X_train, y_train):
    """Train Gradient Boosting model"""
    gb = GradientBoostingRegressor(
        n_estimators=200,
        learning_rate=0.1,
        max_depth=5,
        random_state=42
    )

    gb.fit(X_train, y_train)
    return gb

# Usage
gb_model = train_gradient_boosting(X_train, y_train)

Model Evaluation

Comprehensive Evaluation

from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score
import numpy as np

def evaluate_model(model, X_test, y_test, model_name="Model"):
    """Comprehensive model evaluation"""
    predictions = model.predict(X_test)

    metrics = {
        'MAE': mean_absolute_error(y_test, predictions),
        'RMSE': np.sqrt(mean_squared_error(y_test, predictions)),
        'R²': r2_score(y_test, predictions),
        'MAPE': np.mean(np.abs((y_test - predictions) / y_test)) * 100
    }

    print(f"\n{model_name} Evaluation:")
    print(f"  MAE:  ${metrics['MAE']:,.0f}")
    print(f"  RMSE: ${metrics['RMSE']:,.0f}")
    print(f"  R²:   {metrics['R²']:.3f}")
    print(f"  MAPE: {metrics['MAPE']:.1f}%")

    return metrics, predictions

# Usage
metrics, predictions = evaluate_model(model, X_test, y_test, "Linear Regression")

Compare Multiple Models

def compare_models(models, X_test, y_test):
    """Compare multiple models"""
    results = []

    for name, model in models.items():
        metrics, _ = evaluate_model(model, X_test, y_test, name)
        metrics['Model'] = name
        results.append(metrics)

    comparison = pd.DataFrame(results)
    comparison = comparison.set_index('Model')

    print("\nModel Comparison:")
    print(comparison.round(2))

    return comparison

# Usage
models = {
    'Linear Regression': linear_model,
    'KNN': knn_model,
    'Random Forest': rf_model,
    'Gradient Boosting': gb_model
}
comp