VERIFIED PROJECT RESULTS · AUGUST 2026
From raw datasets to explainable model results.
Four end-to-end Python projects covering exploratory data analysis, classification, regression, clustering, ensembles and recurrent neural-network forecasting.
NumPy • Pandas • scikit-learn • TensorFlow/Keras • Matplotlib • Seaborn
4Completed projects14+Models evaluated100%Reproducible outputs
CLASSIFICATION · REGRESSION · CLUSTERING
Heart Disease Machine-Learning Analysis
Patient records were cleaned, explored, transformed and evaluated through multiple supervised and unsupervised models.
Workflow
Missing-value treatment, descriptive statistics, scaling, train/test split, five classifiers, soft-voting ensemble, two regressors and K-Means cluster selection.
Decision insight
The ensemble produced the strongest discrimination, while the cluster analysis revealed five data-driven patient profiles for exploratory segmentation.
Responsible interpretation
This is an educational data-science demonstration—not a clinical diagnostic system. Medical decisions require validated clinical models and professional review.
HEALTHCARE INFRASTRUCTURE ANALYTICS
USA Hospital Capacity & Helipad Intelligence
National hospital records were used for bed-capacity regression, helipad classification and facility clustering.
Workflow
Data cleaning, hospital-type profiling, capacity modelling, ensemble classification, regression evaluation and K-Means segmentation.
Decision insight
Helipad availability was highly classifiable in the prepared feature space; bed prediction was useful but retained substantial uncertainty.
Responsible interpretation
Facility locations and attributes require current authoritative validation before planning emergency access or healthcare investments.
TIME SERIES · RNN · LSTM · GRU
Healthcare Stock Price Forecasting
Historical daily prices were transformed into sequences and evaluated with recurrent neural networks and an honest baseline comparison.
Workflow
Time-series feature creation, chronological split, scaling, sequence generation, SimpleRNN/LSTM/GRU training, back-testing and ensemble forecasting.
Decision insight
Neural models closely tracked price levels, but directional accuracy remained near 50–52% and the previous-close baseline was highly competitive.
Responsible interpretation
This is a learning demonstration, not financial advice. High price-level R² does not guarantee profitable or reliable directional forecasts.
REGULATORY TIME SERIES · DEEP LEARNING
FDA AI Medical-Device Authorization Forecast
Monthly authorization counts were modelled with seasonal features and three recurrent neural-network architectures.
Workflow
Monthly aggregation, seasonal encoding, rolling features, chronological split, SimpleRNN/LSTM/GRU training and three-month ensemble projection.
Decision insight
The LSTM outperformed the seasonal baseline in this run; the ensemble projected 28, 28 and 27 monthly authorizations after the supplied period.
Responsible interpretation
These are model outputs from a supplied dataset—not official FDA forecasts. Neural-network results can vary slightly between training runs.
COMMON MODEL PIPELINE
Inspect → clean → explore → model → evaluate → explain
- 01 Load and inspect
- 02 Clean and prepare
- 03 Explore patterns
- 04 Train models
- 05 Compare metrics
- 06 Communicate limits