Prediction of Bank Investors using Neural Network in Direct Marketing

Direct marketing in banking is one of the most effective methods of predicting potential investors. Effectiveness of direct marketing is being analyzed using different methods like feature correlation, dataset balancing, neural network (NN) etc.nbsp Usually sixteen to twenty parameters are collected for training database to evaluate the potential client. A fully connected multilayer NN is developed that gradually optimizes the connection based on training dataset. This NN is used to predict the customer willingness for long term deposit with accuracy hire then 95% which corresponds to Accuracy, Sensitivity and Specificity of 95.19%, 92.32% and 95.42% respectively. One of the important parameter is false negative prediction which is 0.63% for above accuracy. Result of false negative indicates incorrectly predicting unwilling clients. With our algorithm, analyzing UCI test benchmark dataset gives 276 true prediction out of 451 records of customers who buy the bank product and only 23 false prediction out of 3668 records of customers who did not buy the bank product. This may be noted that false negative to true negative ratio increases rapidly with small decrease of accuracy. 2% decrease from 95% increases the false negative value fromnbsp 23 to 379. Such increase leads to several fold non productive persuasion effort. On the other hand decrease in true positive reduces the true buyer but do not reduce the productivity due to false prediction. However it is seen that increase of network size do not increase the accuracy even after several hours of training. Hence an optimum size of the network needs to be achieved with automatic iterative pruning