Rainfall-Prediction

Project mission:

This project is developed to predict the rainfall in millimeters in Ramannapeta of Yadadri-Bhuvanagiri district of Telangana.

Point to be noted:

This data set is real time dataset and we have done from data pre-processing to metrics evaluation

Project done by:

Dineshwar Doddapaneni
P.Lakshmi manaswini (https://github.com/LakshmiManaswini-7)

Data Source:

Data is collected from open data websites

Process

Here we used 3 different methods to predict the rainfall,
1)Deep Neural Networks
2)Decision Tree Regressor
3)Random Forest Regressor

Deep Neural Networks:

Here we have used a deep neural network of 4 hidden layers,
After training the data seeing the MSE we thought this were not the best choice(41 in training)

Reasons for Deep Neural Networks Failure:

1. We havent managed to get high quantity of data
   
2. we have trained the machine only with 1000 epochs so due to lack of computational power
3. The architecture was also not sufficent

Decision Tree Regressor

Here we used max depth equal to 5
the R2 score for this model 0.96(train)
the mean absolute error is 0.45(train) 6.9(test)

Random Forest Regressor

here we used Random forest Regressor
the R2 score for this model 0.86(train)
the mean absolute error is 1.029(train) 4.8(test)

• out this we felt Decision Tree Regressor algorithm is best

Improvements that can be done:

• Here we can collect more data and use neurals networks
• more computational power could be really usefull for us