Prediction of Stock Performance Using Deep Neural Network
Achieved a 5.5% earning rate on average for each stock using LSTM Model
Freelance · Shipped · –
ML · Data Analytics · Python · PyTorch
Contract work, replicated the paper https://www.mdpi.com/2076-3417/10/22/8142 in a week, sourced the data from https://data.nasdaq.com/databases/EOD
Used ranking based stock classification to predict top 10% of stocks which will perform well in the next 12 weeks, given the past 52 weeks of data. Result was 5.5% returns on average across stocks predicted
# the aim is this, higher the better
# TNR: 0.975 TPR: 0.041 ACR: 0.508 Precision: 0.154
# saved version one at 40 epochs
# TNR: 0.9797, TPR: 0.0432, ACR: 0.5115, Precision: 0.1235
# Epoch 40, Train Loss: 0.1781, Validation Accuracy: 0.92164156
# saved version two at 50 epochs
# TNR: 0.9822, TPR: 0.0381, ACR: 0.5101, Precision: 0.1240
# Epoch 50, Train Loss: 0.1614, Validation Accuracy: 0.92365932