A tuned Random Forest regression model deployed via a FastAPI backend and React frontend.
applied machine learning / AI House Price Estimator
one inference pipeline four rigorous mathematical boundaries
Follow a California housing record through the inference pipeline. An 8-dimensional census block is normalized, processed through 200 bootstrapped decision trees, cross-validated via GridSearchCV, and served in under 10ms via asynchronous FastAPI.
Scikit-Learn 1.4 • Joblib Serialization • 5-Fold Cross Validation
raw census signalsnormalized 8D vectors
bootstrapped subsets200-tree ensemble vote
36 candidate configurationsR² 0.8062 convergence
HTTP POST requestinstantaneous valuation
01 // SPATIAL INGESTION
Raw census block records are transformed into continuous z-scores across median income, room ratios, household densities, and high-precision GPS coordinates.
02 // TREE BAGGING
Bootstrapped decision trees vote on non-linear price splits, dampening outlier leverage in multi-modal California real estate sub-markets.
03 // GRIDSEARCHCV OPTIMIZATION
Exhaustive grid search isolates optimal tree depth and split criteria across 5 held-out folds, preventing localized data leakage.
04 // FASTAPI SERVING
Serialized model weights are hosted on an asynchronous FastAPI worker with Pydantic validation and interactive Swagger documentation.
Random Forest + FastAPI + React, tuned with GridSearchCV and deployed on Render and Vercel.
Cross-validated coefficient of determination across held-out splits
Mean absolute error on test split (~$32.6k deviation envelope)
Asynchronous FastAPI REST endpoint execution latency
Primary economic predictor of district property values
Proximity to coastal economic hubs (SF Bay Area, Silicon Valley, LA basin)
Density ratio measuring household crowding versus luxury single-family ratios
Physical spatial layout and structural capacity index
Construction era reflecting historical preservation and municipal infrastructure
Minimizes tree correlation & ensemble variance
Captures non-linear geospatial interactions
Eliminates leaf-level memorization of localized noise
Preserves coordinate coupling (latitude + longitude)
React Web Dashboard & Form Controls
Interactive parameter inputs with client-side validation, toast notifications, and Axios HTTP client communicating directly with the FastAPI cloud backend.
Inputs incorporate 8 numerical dimensions from the California housing dataset including median income, housing age, average rooms, population density, and GPS coordinates.
import pandas as pd
import numpy as np
def prepare_features(df):
features = ['MedInc', 'HouseAge', 'AveRooms', 'AveBedrms', 'Population', 'AveOccup', 'Latitude', 'Longitude']
return df[features]The model builds 200 regression trees across bootstrapped samples, averaging splits to minimize variance and prevent overfitting on outlier districts.
from sklearn.ensemble import RandomForestRegressor
model = RandomForestRegressor(
n_estimators=200,
max_depth=None,
min_samples_split=5,
random_state=42,
n_jobs=-1
)Exhaustive hyperparameter sweeps via GridSearchCV with 5-fold cross-validation optimized tree depth, feature splits, and leaf samples to achieve an R² of 0.8062 and MAE of 0.3261.
from sklearn.model_selection import GridSearchCV
param_grid = {'n_estimators': [100, 200], 'max_depth': [20, 30, None]}
grid_search = GridSearchCV(estimator=RandomForestRegressor(), param_grid=param_grid, cv=5, scoring='r2')
grid_search.fit(X_train, y_train)Tuned model weights are serialized using Joblib and loaded into a lightweight FastAPI service, enabling instantaneous predictions through documented Swagger REST endpoints.
from fastapi import FastAPI
import joblib
app = FastAPI()
model = joblib.load("random_forest_model.joblib")
@app.post("/predict")
def predict_price(data: HouseFeatures):
prediction = model.predict([data.to_array()])
return {"predicted_price": float(prediction[0])}