Improving the Trust Factor in Decentralized Networks through Deep Learning

Recent developments in techniques and methods of Machine Learning have led to enhanced techniques of predictive modeling. This research focuses on the performance of the four commonly used models as CNN, RNN, ANN, LSTM and studies them based on the parameters of accuracy, precision, recall, F1 score. In the introduction, we emphasize the relevance of choosing the best model when working with such problems as forecasting needs an accurate and reliable result. The experience of the prior publications points out that, yet CNNs and ANNs dominate the field, RNNs and LSTMs turn into better options when it comes to temporal relations and patterns recognition. In our case, we used a comparative analysis in which a set of bar charts and learning curves was used to compare each model’s effectiveness. Bar graphs allowed comparing metrics of the CNN, RNN, ANN, and LSTM methods; learning curves were used to observe the training of models during 20 epochs. It is also visible that the RNN model has the highest accuracy (97. 96%) and precision (98. 00%), as well as a high recall (97. 50%). The LSTM model follows with the accuracy of 97. 63% and an F1 score of 97. CNN and ANN results are relatively poor with CNN with accuracy of 96%. 89% though ANN was higher and stands at 95. 89%. The authors’ investigation finds the comparative analysis under consideration proves RNNs to be the most resistant in relation to the parameters studied, which makes this approach exceptionally suitable for precisely targeted problems as to precision and recall. This enables LSTM models to give a fairly good balance, with CNN as well as ANNs being only slightly behind the top performing models. These findings highlight the relevance of the problem of model selection for achieving the best predictive accuracy.