Time Series for Univariate Dataset using RNN with LSTM

Here I have implemented RNN with LSTM for Univariate Time Series dataset from one of the kaggle's competition:

https://www.kaggle.com/c/predice-el-futuro/overview

Problem Discription:

Data consist of two files:

• train_csv.csv: The training dataset consists of 80 observation with observation id, timestamp and feature value.

• test_csv.csv: The next 40 observation along with observation id and timestamp are present here. You'll have to predict the value of feature at these timestamps.

• value to predict: The value of feature attribute has to be predicted.

File Description:

• train_csv.csv - the training set

• test_csv.csv - the test set

---

Here I have implemeted RNN with LSTM with loss value lesser than 0.03 while training and RMSE value 122.90 on validation split.

Loss Function:

Prediction Graph:

• Blue : Actual Value
• Orange : Prediction on Training sets
• Green : Prediction on Validation sets