hospital-lstm-forecast

Hospital Demand Forecasting (LSTM Baseline)

A model for predicting demand for medical services/products based on historical data.

Description

That LSTM-model (Long Short-Term Memory), trained on a dataset Monash Time Series (Hospital). The dataset contains 767 monthly time series for the period 2000-2006yy.

Task: Predict the value of the next Установка зависимостейmonth based on the previous 12 months.

Architecture

• Type: Deep Learning (LSTM)
• Framework: TensorFlow / Keras 3
• Layers:
  • LSTM (64 units, return_sequences=True)
  • Dropout (0.2)
  • LSTM (32 units)Установка зависимостей
  • Dropout (0.2)
  • Dense (1 unit)
• Training: Adam optimizer, MSE loss.

Limits (Baseline)

This version usr a Global Normalization.

• The means that the model works best on averaged data.
• For series with low amplitude or outliers, the model may produce a forecast close to the average value across the entire dataset ("straight line").
• For production, it is recommended to implement order normalization (Per-series normalization).

How to use

1. Installing dependencies

pip install tensorflow keras huggingface_hub

### 2. Download and forecast
from tensorflow.keras.models import load_model
import numpy as np

# to load model
model = load_model("hospital_lstm_model.keras")

# Example of input data (12 normalized values [0..1])
# IMPORTANT: To get the result, you need to apply the same normalization as during training
input_data = np.random.rand(1, 12, 1).astype(np.float32)

prediction = model.predict(input_data)
print(f"Forecast (normalized): {prediction[0][0]:.4f}")

The forecast_visualization.png file shows an example of the model's operation
(Red line is the forecast, Green line is the reality)


### Author
Created for educational purposes.