Rain Predictor Neural Network

A simple neural network written entirely from scratch in Python that predicts the chance of rain using only:

• Humidity
• Cloud cover

The model trains itself using backpropagation, saves the best weights, and allows you to test predictions later.

It also includes premade datasets for:

• Seattle (April–May)
• Bangalore (April–May)

---

Features

• 2-input neural network
• 4 hidden neurons + 1 output neuron
• Sigmoid activation function
• Xavier weight initialisation
• Automatic learning rate decay
• Automatic restart if training gets stuck
• Saves the best model as a .pkl file
• Separate model loader and tester
• Premade Seattle and Bangalore weather datasets

---

How It Works

Inputs:

• Humidity (0.00–1.00)
• Cloud cover (0.00–1.00)

Output:

• Rain probability from 0.00–1.00

Example:

Humidity	Cloud Cover	Predicted Rain Chance
0.45	0.12	Very unlikely
0.65	0.80	Likely
0.74	0.78	Very likely

---

Neural Network Structure

2 Inputs
   ↓
4 Hidden Neurons
   ↓
1 Output Neuron