Investigating the Impact of Feedback Loop Training on Machine Learning Model's Performance in Time Series Data

Ever since the development of Artificial Intelligence (AI) and Machine Learning (ML), the need for raw data has increased to mitigate this problem we investigate the impact of feedback loop training on an LSTM model's time series data and the performance of the model is investigated. We also employ traditional ML models like RandomForest Regression to perform a comparative study with the LSTM model. We seek to examine the effect on the accuracy and convergence behavior of the model by using its output as input for further iterations. A climate time-series dataset from Kaggle is used to train the LSTM model iteratively over ten iterations. According to our research, feedback loop training can have benefits like capturing intricate dependencies, but the cost outweighs the results as the model accuracy decreases drastically over several iterations of training. We have also compared the Model performance with traditional Machine Learning algorithms and other types of Neural Networks like ANNs.