Next purchase brand prediction using lstm using keras library. Here i have formulated a dataset of customers having txn for different brands,then using lstm,customers previous txn history, have predicted next n brands customer is likely to transact.
Got Precision: 0.010211932550116815, Recall: 0.01240272373540856, F1 Score: 0.008605248175606817.
My observation is since the data was formulated randomly, the model did not work good and the precision is low. With real data, it could perform much better. Also hit rate will also be better. Also , the quantity of data was less, more data can help.Also i was working on google colab so computation resources i had was limited. Hyperparameter tuning could have helped. Also, using item features and customer features , with some different model approach can give us better results.