shyletnyazika/healthconnect-experience-lab

AnalystLab Africa Experience Lab — improving patient appointment attendance using data and AI (Data Science track) # HealthConnect Experience Lab — Week 4 (Data Science Track) **AnalystLab Africa — Experience Lab Internship Programme** ## Objective Define the machine learning problem for predicting patient appointment no-shows at HealthConnect Clinic, and assess whether the provided appointment data can realistically support that goal. Week 4 is a foundation stage — no model is trained or deployed at this point. ## Business Context HealthConnect Clinic experiences missed appointments (no-shows), which waste appointment slots and disrupt care. The central project question: *how can HealthConnect Clinic use data and AI to reduce missed appointments and improve the patient support experience?* This track's contribution is the Data Science / predictive modelling angle. ## Files in This Submission | File | Description | |---|---| | `HealthConnect_Week4_Notebook.ipynb` | Data assessment, target definition, and initial modelling plan | | `HealthConnect_Week4_ML_Problem_Definition.docx` | Full problem definition, data assessment, and modelling approach | | `Week4_Project_Summary.docx` | Concise summary of Week 4 work and Week 5 focus | ## Proposed ML Task Binary classification: predict `no_show_target` (1 = No-Show, 0 = Attended). Cancelled appointments (5.3% of data) are excluded from this first model, since a cancellation is a communicated decision rather than an unexplained absence. ## Key Findings - 5,000 appointment records, 18 variables, no duplicates. - Outcome split: No-Show 48.5%, Attended 46.3%, Cancelled 5.3%. - `reminder_channel` missingness (27.3%) is structural — it corresponds exactly to appointments where no reminder was sent, not a data quality issue. - No-show rate among non-cancelled appointments (~51.2%) is close to balanced, simplifying the initial classification approach. ## Key Risks Identified - Possible data leakage via `waiting_time_minutes` (may only be known after the appointment). - Repeated `patient_id` values across appointments — random train/test …