California_House-Price-Prediction

This is my first end-to-end ML project implementation covering all required stages taking guidance from the book called "Hands-On Machine Learning"

Table of Contents

• Big Picture
• No Data Snooping
• Exploratory Data Analysis
• Data Preparation
• Model Selection and Training
• Model Fine Tuning
• Evaluation on Test Set

Big Picture

1. Problem Statement

   • Welcome to Machine Learning Housing Corporation!
   • Organization Objective: Replacing expensive, time-consuming and less effective manual prediction techniques with Machine Learning

2. Framing the problem

   • Data: California Census Data
   • A typical Univariate Multiple Regression task.
   • Training Set is labelled hence "Supervised Learning"
   • Data is small hence we shall opt for "Batch Learning"

3. Performance Measure

   • Root Mean Square Error (RMSE) - l ₂ Norm
   • Mean Absolute Percentage Error (MAPE)

No Data Snooping

1. Get the Data
   • Overview and Primary Understanding
2. Test Set
   • Firstly employed Simple Random Sampling to draw a test & train set using Scikit- Learn
   • Secondly utilized Stratified Sampling by categorizing the whole datase on median_income
   • Later we compared Sampling Bias from both the sampling techniques

Exploratory Data Analysis

1. Creating Viualizations

---

---

• There is a celar depiction of clusters in and around San Diego, Los Angeles, San Feancisco, etc.
• From the above figure we can see a general figure that ocean_proximity seems to be associated with median_price_value
• Bus still there are exceptions in North California, so we've to deploy some feature engineering here as well.
• Features such as proximity to clusters can also be checked.

---

2. Correlation Matrix and Scatter Plot

---

• In General it shows a strong positive trend.
• Straight line at $500,000 reemphasize the price_cap
• Concerns are a few straight lines in and around $450,000, $350,000, $280,000, $230,000 and so on in the below.
• We may remove the concerned districts.

Data Preparation

1. Data Cleaning

   • Missing value of total_bedrooms (1.01%) has been treated using SimpleImputer Class of Scikit Learn.

---

---

2. Handling Text Attribute

   • OneHotEncoding is used to handle ocean_proximity column

3. Feature Scaling and Transformation Pipeline

   • Laid down a single transformation pipeline to transform both numeric and categorical attributes
   • StandardScaler(), ColumnTransformer() classes have been utilized.

Model Selection and Training

1. Model Selection

   • Linear Regression, Decision Tree Regressor, and Random Forest Regressor models have been fitted on the training set.

2. Model Evaluation

   • "K-Fold Cross Validation" depicts RSME scores of 68973.97, 69919.68, 50631.51 respectively.
   • Random Forest Regressor looks very promising.
   • Note: The score on the training set is still much lower than on validation sets, which means still overfitting the training set.

Model Fine Tuning

1. Grid Search Cross Validation

   • Deployed GridSearchCV() to fine-tune hyperparameters
   • Got ---> RandomForestRegressor(max_features= 6, n_estimators= 30)
   • RSME score slightly improved from 50631.51 to 50586.27 (K Fold Cross Validation)

---

---

Evaluation of Test Set

• K-Fold Cross Validation (RSME) score of 46811.29 having [44833.99542182, 48708.38642949] confidence interval with a 5% level of Significance.
• Mean Absolute Percetage Error(MAPE) score 0.1771.

---

🛡️ Demonstration Video

---

Thank You So Much...🙏🙏