Published Paper Details:

Machine learning has an obligatory reliance on causal inference which is actually the search for cause and effect relationships in observational data, that is crucial in a variety of fields like economics, healthcare, and social sciences. Traditional methods for causal inference often face challenges with high-dimensional information and interactions between variables that are complex, making such techniques unavailable for biased or imprecise estimates of treatment effects. This research presents a new method which links deep learning with instrumental variables to solve these problems thereby providing a strong empirical approach for causal inference in machine learning. Our study builds a deep learning model that captures intricate and nonlinear association among variables while utilizing instrumental variables to control unobserved confounding factors. Synthetic datasets as well as actual clinical data were used to validate our proposed approach. A simulated dataset was used where there were pre-defined causal relationships in a healthcare setting to enable examination of the validity under conditions strictly controlled by the investigator. Realistic public health and economic datasets have been used to illustrate the feasibility of our method. The outcomes of the experiment show that the application of our method increases the reveal of the real ATE and CATE comparing to the traditional approach. MSE for example and bias are some of the aspects that are minimized by the deep learning model hence provide a reliable chance or causal inference. It shows the effects of the treatment while in the case of more elaborate analysis, attention is given to the ability of the treatment to affect different groups. Based on these findings, it is possible to consider this new approach helpful for management of observational data, which are complicated in their nature; the proposed approach is useful for providing professionals of various fields with necessary information. This paper presents a new approach developed to eradicate any constraint observed in traditional procedures particularly when performed under high dimensionality; information that enlarges the understanding of the topic area. The interpretation skills of the model need improvement in the future, as well as its ability to autonomously choose relevant instruments and apply them in different fields.It is worth pointing out that this particular research helps us understand how causality works within complicated data composition, especially in machine-based inference among others