DIIT: A General Model for Time Series Projections, Proven on NIFTY Index Funds

To know the future is to know the past. The ability to properly estimate the future of a system is an elusive problem. Researchers have developed many tools to do just that, but a unified approach does not exist. Intertemporal causalities are main signages for predictions in computational finance. Here, since past value of a variable is highly correlated with the present and future of that variable, time series data analytics is much sought after modality for predictions. For a large temporal data set, time period bias is a very common sampling error, resulting in circumstance-specific unique observations only. Experts cannot extend such observations to a larger industry with wider problem spaces. In this paper, we propose a solution to fit any time series data, with an aim to eliminate the time period bias. In this work, we have created a system that meshes previously created systems such as ARIMA, ARMA, and AR. This helps to create a dynamic system that conforms to the specified time series data and modulates to create a specialized architecture for future prediction. We have taken test cases with varying hyperparameters and found a median accuracy of 94.95 % with a minimum delay in the training of 7 days and a median delay in training the model of 60 days.