GIVE YOUR BEST PRICE THE ONE I WENT WITH IS NOT REAL (10-11) Project: **A Flask + MySQL web system with Machine learning module for preliminary classification and statistical analysis of suspicious death cases in Albania, with expert review and correction.** Use only: * Python Flask * MySQL * HTML/CSS/Bootstrap 5 * Chart.js * SQLAlchemy * Flask-Login * Flask-WTF or equivalent * Werkzeug security * scikit-learn * pandas, numpy * joblib Important requirements: * Everything must be built **from scratch** * No public/premade dataset * No pretrained/premade model I need an **original synthetic dataset** created from scratch only for this project. It will be temporary and later replaceable with real data. Dataset requirements: * generated with custom Python code * realistic but simple forensic-style patterns * around **1500–3000 rows** * some missing values * some noisy/ambiguous/contradictory cases * labels generated logically/probabilistically, not randomly Required outputs: * `dataset/cases_synthetic.csv` * `dataset/data_dictionary.md` * `app/ml/generate_synthetic_dataset.py` Important: logical/probabilistic rules may be used only to **generate the synthetic dataset**. The **final prediction in the app must come from ML models trained from scratch**, not only rules. Train and compare only **2–3 ML models**, for example: * Logistic Regression * Decision Tree * Random Forest Choose the final model after: * preprocessing * feature engineering * train/test split * basic cross-validation * simple hyperparameter tuning * evaluation This should be a **decision support system**. Flow: * user enters case data * system stores it in MySQL * model gives preliminary prediction * system shows predicted class and confidence * expert reviews it * expert confirms or changes final class * system stores both model prediction and final expert decision Classes: * Suspected homicide * Suspected suicide * Suspected accident * Suspicious natural death * Undetermined cause Roles: * **Admin**: manage users, cases, dashboard * **Operator/Expert**: create/edit cases, view prediction, assign final class, add expert comment Core features: * login/logout * role-based access * full CRUD for cases * search/filter * dashboard with charts * ML prediction module * case details page * expert correction workflow * basic audit log Use mostly **structured fields**, not much free text. Include fields such as: * case code * case date * age * gender * region * scene/location * injury type * violence indicators * toxicology * medical history * case status * priority * some trauma indicators * some medical/psychiatric indicators * weapon/vehicle/suicide note/witness/conflict flags Use free text only where needed, such as short case description and expert comment. Before prediction include: * input validation * missing critical field check * contradiction/inconsistency flags * basic input quality score ML pipeline should include: * data loading * cleaning * missing value handling * feature engineering * encoding * split * training * evaluation * model saving * Flask integration Use proper scikit-learn components such as: * Pipeline * ColumnTransformer * SimpleImputer * OneHotEncoder Evaluate with: * accuracy * precision * recall * F1-score * confusion matrix Also include: * confidence score * low-confidence flag for uncertain cases Prediction output should store/show: * predicted_class * confidence_score * low_confidence_flag * prediction_date Prevent **data leakage**. Do not train on fields known only after expert review, such as: * final_class * expert_comment * final decision timestamps Database tables should include at least: * users * cases * predictions * final_classifications * audit_logs Dashboard/statistics should include: * total cases * cases by region * cases by gender * cases by category * cases over time * high-priority cases * model vs expert agreement rate Security requirements: * secure password hashing * proper session management * strict role-based access control * server-side validation * CSRF protection * use SQLAlchemy queries only * safe handling of secrets through environment variables * safe error handling without exposing internal details Project structure should be clean and modular, for example: * app/ * models/ * routes/ * forms/ * services/ * ml/ * templates/ * static/ * migrations/ * dataset/ * trained_models/ * tests/ * run.py * requirements.txt * README.md I also need: * SQL schema * SQLAlchemy models * Flask routes * Bootstrap templates * synthetic dataset generator * ML training script * prediction integration * testing * local run instructions Build it in this order: 1. project structure 2. database schema 3. models 4. routes 5. templates 6. synthetic dataset generation 7. preprocessing 8. model training 9. evaluation 10. model saving/loading 11. Flask integration 12. dashboard/statistics 13. testing 14. final run instructions Finally, create **`TEXT.md`** with clear step-by-step instructions for: * project structure * synthetic dataset generation * label generation logic * preprocessing * feature engineering * ML training and model selection * Flask prediction integration * local setup and running * retraining from scratch * replacing synthetic data with real data * testing * explanation of the main files/folders * implemented security measures The instructions in `TEXT.md` must be clear, practical, and simple enough for a bachelor student to follow.