Data Science: Diabetes Prediction- Model Building Deployment

Explore the diabetes prediction problem by loading a dataset in a Jupyter notebook, understanding and cleaning data, preparing features like BMI and insulin, and evaluating multiple models for outcome prediction.

Load data from the source into a data frame, read a csv, and inspect rows to confirm features including pregnancy, glucose, blood pressure, skin thickness, insulin, bmi, and outcome.

Explore data quickly with Pandas Profiling: generate a one-command profile report that reveals dataset size, missing values, distributions, correlations, and sample rows for rapid understanding.

Explore data structure with pandas profiling, inspect 768 rows and 9 features, check for nulls, describe numerical fields, and plot a diabetic vs non-diabetic pie chart using the target variable.

Identify zero values in key features and replace them with missing data. Impute by medians per outcome and compute missingness to support diabetes prediction.

Perform a 70/30 train-test split using the x features and y outcome death, with random_state 42 and stratified sampling to preserve equal zero/one distribution, preparing for model training and deployment.

Scale features using a standard scaler to bring values to the same scale, speeding model training and enabling deployment on unseen data, with confusion matrix and classification report for evaluation.

Explore the confusion matrix in diabetes prediction, learning how true positives, false positives, true negatives, and false negatives reveal model performance and inform the classification report.

Explore how the classification report extends the confusion matrix by detailing precision, recall, and support. Calculate precision and recall manually and examine their tradeoff for diabetes prediction.

Explore how the ROC curve plots true positive rate against false positive rate across thresholds, and how the AUC quantifies a classifier’s ability to distinguish positive from negative classes.

Compare models on the diabetes dataset with five-fold cross-validation to identify the best, visualize results with a box plot, and highlight random forest as top performer before hyperparameter tuning.

Select the random forest model with default parameters after it outperformed other models. Create a random forest classifier, feed data to predict, and review classification reports to assess its performance.

Apply a random forest classifier to diabetes data and evaluate performance. Use a classification report, confusion matrix, and ROC-AUC to interpret accuracy and test predictions.

Explore hyperparameter tuning with randomized search CV for a random forest classifier, comparing to grid search CV and optimizing accuracy under computational constraints.

Create the final random forest classifier using the selected hyperparameters from tuning, fit it on the training data, and compare its predictive performance to the vanilla model.

Assess the final diabetes prediction model with classification reports, confusion matrices, and AUC-ROC, reporting class accuracy and overall performance to inform deployment.

Explore final inference of a random forest diabetes prediction model, compare training fit with tuned variants, and save the final model for later deployment and cloud-based inference.

Load saved models and use them for predictions on unseen data, avoiding re-training. Store model weights in a saved model to enable faster, flexible deployment for future diabetes predictions.

Test the diabetes prediction model on unseen data by creating random data, applying scaling with the scalar, predicting with the model, and preparing for cloud deployment.

Explore streamlit as an open-source python library to quickly build interactive web apps for showcasing machine learning models and deployment, with minimal front-end effort and professional visuals.

Create a frontend interface to interact with the trained model, load the model from disk, preprocess inputs with a scaler, run inference, and display health status in the user interface.

Run the streamlit app locally with streamlit run app.py to launch the UI and predict diabetes from glucose, skin thickness, blood pressure, insulin, and bmi.

Deploy a diabetes prediction model to the Heroku cloud by creating an app in Europe region, linking a GitHub repo, and manual deployment, then test predictions via a public URL.