Overview

In this work, I aim to span a various Machine Learning models; especially the regression ones. And the goal is to select the most promissing model that can predict in a smart fashion the Atomization energy of a given molecular structure. The Data can be found in the DataBase called GDB−13, also it is available under the name roboBohr as csv file format in kaggle’s Dataset energy-molecule. The idea was a pivot of discussion in several works [M. Rupp et al , Lorenz C. Blum and J-L Reymond and lately Burak Himmetoglu ]. Moreover, the present notebook consist of learning about the suitable models that can be deployed; and to investigate how far I can achieve in term of performance. I will start as usual, by pre-processing the data (cleaning, droping NAN values … etc), then I tackle very interesting section in data transformation fields mainly the so called Singular Value Decomposition (SVD). After that, I begin testing the models and norrowing the bands of these smart objects. The chosen models will undergo a testing performance according to various scores metrics. An adequate way for biasing the flow work is cited. A comparison between some strategies will be highlighted. After that, I try to tune the Hyperparamaters, and disscusse a features engineering attempt.