Multivariate Time Series Forecasting to Forecast Weight Dynamics

This paper represents a multivariate time series forecasting approach to predict the weight dynamics of an insect colony. In this paper, we are predicting the Colony Weight of the Insect which has the main impact on the growth and overall health of the insects. To accomplish this task, we have performed extensive preprocessing on the time series data, ensuring data quality and suitability for analysis. Overall, we have conducted a high-level Exploratory Data Analysis which covers time series analysis to identify different patterns as well as trends present in the time series data. Following the EDA, Feature Engineering and Feature Selection techniques have also been performed to find the features that have an impact on the colony weight feature. As a follow-up, we have performed Multivariate Time Series forecasting of the colony weight for future data with Time Series Forecasting techniques such as Vector Autoregressive moving average with exogenous regressors (VARMAX) and with Deep Learning techniques such as Long Short-Term Memory (LSTM). Several statistical parameters such as AIC score, BIC score as well as MSE value along with the grid search method have been utilized to find the optimal lag value which affects the model performance significantly. The performance of the black box model has been compared with the interpretable model to find the best outcome. This paper contributes to the field of multivariate time series forecasting in the context of insect colony dynamics. Various insights that have been gained from that study have implications for understanding insect behaviour, population dynamics, and overall ecosystem health. Furthermore, the proposed methodology can be extended to other domains involving multivariate time series forecasting.