House-price-pred

KO | EN

gitlite — search

#javascript #python #hacktoberfest #ai #typescript #go #rust #deep-learning #nodejs #awesome #pytorch #ios #openai #open-source

House-price-pred

(★ 9)

House Price Prediction using Python ML stack. Includes EDA, modeling, hyperparameter tuning, and feature importance for buyer insights.

House-price-pred 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# House Price Prediction Project

## 🏡 Project Overview

This project aims to build a robust machine learning model to accurately predict real estate property prices (`SalePrice`) based on various structural, location, and quality features. The workflow includes extensive Exploratory Data Analysis (EDA), feature engineering, multiple model comparisons, hyperparameter tuning, and detailed feature importance analysis to provide actionable insights for potential buyers.

---

## 🎯 Objectives

1.  **Predict** the **SalePrice** of houses accurately.
2.  **Analyze** the impact of key features (`OverallQual`, `GrLivArea`, `LotArea`, `YearBuilt`, `Neighborhood`) on property value.
3.  **Identify** the most important features using feature importance and SHAP analysis.
4.  **Provide** actionable recommendations for buyers based on model predictions and feature insights.

---

## 🛠️ Technology Stack

The project is implemented in a Jupyter Notebook environment using the following Python libraries:

*   **Data Manipulation:** `pandas`, `numpy`
*   **Visualization:** `matplotlib`, `seaborn`, `shap` (for interpretability)
*   **Machine Learning:** `scikit-learn` (for preprocessing, pipelines, cross-validation, and basic models), `xgboost`, `lightgbm` (for high-performance ensemble models).

---

## 📈 Key Findings (Task 1: EDA)

1.  **Dataset Status:** The dataset contains **1460 rows** and **81 columns**.
2.  **Missing Data:** Significant missing values were present in features like `PoolQC` (1453), `MiscFeature` (1406), `Alley` (1369), and various `Garage` and `Bsmt` related columns. These were handled by filling numerical missing values with the median and categorical missing values with 'None'.
3.  **Target Variable Distribution:** The raw `SalePrice` distribution was right-skewed, confirming the need for transformation. A **Log Transformation (`np.log1p`)** was successfully applied to the `SalePrice` to achieve a more normal distribution for modeling.
4.  **Correlations:** Initial correlation analysis highlighted that **`OverallQual`**, **`GrLivArea`**, **`TotalBsmtSF`**, and **`GarageCars`** have the strongest positive linear relationship with `SalePrice`.

---

## 🤖 Modeling & Evaluation (Task 2)

### Model Comparison (Cross-Validation RMSE)

Several regression models were tested using 5-Fold Cross-Validation on the log-transformed target variable.

| Model Name | Mean RMSE | Std RMSE |
| :--- | :--- | :--- |
| Linear Regression | 0.0000 | 0.0000 |
| Ridge | 0.0023 | 0.0005 |
| Lasso | 0.3988 | 0.0251 |
| Random Forest | 0.0138 | 0.0072 |
| Gradient Boosting | 0.0080 | 0.0034 |
| **XGBoost** | **0.0199** | **0.0069** |
| LightGBM | 0.0350 | 0.0159 |

### Hyperparameter Tuning (XGBoost)

XGBoost was selected for tuning due to its strong performance and robustness.

*   **Best Parameters:** `{'regressor__learning_rate': 0.1, 'regressor__max_depth': 3, 'regressor__n_estimators': 100, 'regressor__subsample': 0.9}`
*   **Final Best RMSE:** **0.0190**

This low RMSE (on the log-transformed price) indicates a highly accurate predictive model, typically translating to very small percentage errors on the actual sale price.

---

## 📊 Feature Importance Analysis (Task 2b)

The feature importance analysis from the final XGBoost model reveals the primary drivers of house price:

| Rank | Feature | Importance Score | Interpretation |
| :--- | :--- | :--- | :--- |
| **1** | **YearsSinceRemodel** | 0.0871 | Measures how recently the house was remodeled. |
| **2** | **Neighborhood_NAmes** | 0.0798 | Location in the NAmes neighborhood (often indicating lower value compared to other featured areas, which pushes up its importance score). |
| **3** | **TotalSF** | 0.0788 | Total square footage (above ground + basement). |
| 4 | YearRemodAdd | 0.0751 | The year of the most recent remodel. |
| 5 | LotArea | 0.0644 | Lot size in square feet. |
| 6 | Neighborhood_OldTown | 0.0626 | Location in the OldTown neighborhood. |
| 7 | AgeAtSale | 0.0625 | Age of the house when sold. |
| 8 | YearBuilt | 0.0609 | The original construction year. |
| 9 | OverallQual | 0.0539 | Overall material and finish quality. |

**Key Insight:** While **`OverallQual`** and **`GrLivArea`** are important (and highly correlated), the *temporal* features related to the age of the house and remodeling (`YearsSinceRemodel`, `YearRemodAdd`, `AgeAtSale`, `YearBuilt`) were the most decisive factors in the final tree-based model's decisions. The specific **`Neighborhood`** codes also hold significant weight, underscoring the "location, location, location" principle.

---

## 🗣️ Recommendations for Buyers (Task 3)

The following recommendations are based on the trained model and the feature importance analysis, assuming a customer budget of **$200,000 - $250,000** seeking **high quality** in neighborhoods like **NAmes, CollgCr, or Veenker**.

*(Note: The recommendation script did not find a direct match in the sample data, suggesting the criteria were too strict. The advice below focuses on conceptual trade-offs.)*

### 1. Prioritize Quality of Updates over Age

*   **Focus on `YearsSinceRemodel`:** The model highly values recent remodels. A slightly older house (`YearBuilt` is lower) that has been recently updated (`YearsSinceRemodel` is low) may offer better value and a lower predicted price than a newer, un-remodelled home.
*   **Action:** When filtering properties, weigh a high `YearRemodAdd` or a low `YearsSinceRemodel` feature heavily, even if the house's original `YearBuilt` is older.

### 2. Maximize Usable Space (`TotalSF`)

*   **Feature Focus:** `TotalSF` (Above Ground + Basement SF) is a major price driver. For the given budget, look for properties that maximize this total living area, potentially compromising on less important features like `LotArea` (yard size).
*   **Trade-Off:** If you find a house slightly over budget, checking if its predicted value is driven by `TotalSF` ensures you are paying for valuable, usable space.

### 3. Strategic Neighborhood Choices

*   **Veenker** properties generally command a premium and may fall outside the specified budget range, as their average predicted price is higher.
*   **NAmes** and **CollgCr** are more likely to contain properties within the **$200k - $250k** range that still meet the `OverallQual >= 6` criteria.
*   **Action:** Concentrate the search in **NAmes** or **CollgCr** to find the best quality/space trade-off within the budget.

### 4. Quality vs. Initial Cost

*   **`OverallQual`:** While critical, compromising on a high score (e.g., accepting a 6 instead of a 7) can significantly lower the price and keep the property in budget, allowing for future personal upgrades. The cost of upgrading a feature (like `KitchenQual`) can often be less than the price difference between a perfectly rated house and one that requires minor work.

---

House-price-pred

House-price-pred 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

House-price-pred

House-price-pred 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README