Forecasting customer profitability using ensembles of RFM, LSTM and GA

Customer profitability is crucial for enterprises to identify opportunities and conduct effective customer-centered marketing strategies. In the current literature, it is often used for recognizing different deal-sensitive levels of customers and is common in customer relationship management. To capture its dynamic change, this research develops a two-phase ensemble of Recency, Frequency, and Monetary (RFM) model, Long-Short-Term Memory (LSTM) neural network, and Genetic Algorithm (GA). In the first phase, several clusters are identified by integrating the RFM model with K-means method. In the second phase, the number of clients is predicted in each cluster using a GA-LSTM model where the GA optimizes the structure of LSTM model. The experimental results demonstrate that customer segments with different characteristics can be identified. The proposed GA-LSTM model can be used for finding optimal parameters under certain conditions and improving prediction robustness.