Data Science: Car Price Prediction-Model Building Deployment

Create a data project with data and notebook folders, import key libraries (pandas, numpy, seaborn, scikit-learn, joblib), and prepare for data understanding, cleaning, and model building.

Learn to load car price data from an Excel file, explore the data dictionary and column meanings, and preview the dataset in a pandas-powered Jupyter notebook.

Explore the car price dataset by inspecting its shape, data types, and missing values; examine distributions and confirm zero missing values in the data frame.

Identify duplicates by counting unique car ids, drop the car id and car name as irrelevancies, and observe that the model performs better without the symbolic variable.

Perform univariate analysis by examining one variable at a time to understand its distribution, and note the move toward multivariate analysis across multiple variables.

Explore bivariate analysis of car price by body categories using pivot tables and bar graphs, computing mean prices per category and visualizing annotated results.

Explore data binning, also called discrete binning or bucketing, a technique that groups values into defined bins to transform a numerical variable into a categorical variable.

Apply data binning to create a peak r.p.m. bucket using cut into numeric ranges, then build a pivot table and bar chart to compare mean price by bin.

Explore how correlation shows how variables move together and identify features that strongly affect price. Visualize these relationships with a seaborn heat map to spot strong correlations.

Explore plotting scatter plots to visualize relationships between car price and features like engine size, horsepower, and wheelbase. Discover strong correlations through side-by-side visuals and heatmaps.

Visualize how car price and features are distributed using box plots and distribution plots with seaborn, highlighting outliers, skewness, and category-wise comparisons.

Understand outliers as observations with abnormal distance; don't assume removal, as business context matters. Use percentile analysis to detect significant deviations and preserve data in small datasets.

See how one hot encoding converts categorical features into numerical inputs for machine learning models. The lecture shows using a one hot encoder to transform data for car price prediction.

Split the data into x (features) and y (price), then create train/test sets at 70/30 with random_state 42, divide 30 percent into validation and test sets 50/50, and note scaling.

Learn how scaling transforms feature ranges without changing distribution, using min-max and standard scalers, applied to numerical features only and fit on the train set for deployment.

Explore boosting as an ensemble technique that adds models to correct previous errors by predicting residuals, and learn about gradient boosting and XGBoost's speed and performance.

Create an XGBoost regression model with default parameters, fit on training set, predict test set, and report an R2 around 95% and RMSE about 1775.33, preparing for hyperparameter tuning.

Optimize the car price prediction model by tuning hyperparameters with RandomizedSearchCV, using cross-validation and exploring options like learning rate, gamma, and max depth.

Tune an xgb regression model with randomized search cross-validation, compare with default parameters, achieve an R2 score around 95%, and save the tuned model for deployment.

Evaluate the model performance on training and test sets using the R2 score, achieving around 98% and 95%, and interpret the scatterplot of actual versus predicted values to gauge significance.

Plot a scatter plot of actual versus predicted car prices to evaluate model performance, showing points close to the diagonal line and highlighting key features that drive price predictions.

Identify the most influential features for car price prediction using feature importance and coefficients, build and rename a feature-coefficient data frame, then filter and rank positive contributors for deployment.

Learn how Streamlit enables quick creation of interactive machine learning web apps, with a simple installation, localhost deployment, and writing a minimal Python frontend.

Create a streamlit web interface to interact with the trained car price model, loading scaler, one-hot encoder, and model, then predict price using UI controls.

Run the car price predictor locally by executing streamlit run app.py to launch a Streamlit server at localhost:8501, load models and one-hot encoding, and interactively predict prices from feature inputs.

Configure gitignore to exclude ipynb_checkpoints, add setup.sh and Procfile for deployment, and update the readme. Document car price prediction workflow with xgboost and streamlit, including data, models, and deployment notes.

Push your car price prediction project to a GitHub repository, clone and add, commit, and push files to master, then deploy the code to Heroku.

Deploy a car price prediction app to the Heroku cloud by connecting GitHub, deploying the master branch, and exposing a free URL, showcasing Streamlit interfaces and XGBoost models.