A machine learning approach for learning temporal point process

Despite a vast application of temporal point processes in infectious disease diffusion forecasting, ecommerce, traffic prediction, preventive maintenance, etc, there is no significant development improving the simulation and prediction real-world environments. With this problem at hand, we propose novel methodology for learning based on one-dimensional numerical integration techniques. These techniques are used linearising negative maximum likelihood (neML) function enabling backpropagation neML derivatives. Our approach tested two real-life datasets. Firstly, high frequency process data, (prediction highway traffic) secondly, very low dataset, ski injuries resorts). Four different baseline models were compared: second-order Polynomial inhomogeneous process, Hawkes with exponential kernel, Gaussian Poisson process. The results show ability proposed to generalize datasets illustrate how mathematical influence quality obtained models. presented not limited these can be further optimize predict other that processes.