This paper indicates a comprehensive analysis ofResidual Network architecture for better and efficient optimiza-tion. Our main aim was not focused on better accuracy ratherour goal was to design a recommendation system that caneasily rank a network from best to worst case for a betterefficient network performance. In quest of achieving our objectiveseveral activation function, depth-width and several connectionsof a ResNet structure was investigated. Finally, a comprehensiverecommendation was delivered, which even has the scope offurther improvement. For this paper, only two types of activationfunction was tested, but in future, it is possible to make a databasebased on the user recommendation and preference. A note-able finding from our research included that increasing widthenhanced the testing accuracy significantly while an incrementin depth did not have any substantial effect. This discrepancywas not expected and in future, it is conceivable that properregularization mechanism could possibly enhance the bottleneck.Moreover, it is noteworthy that, our goal of this project was noton comparing our results with state of the art accuracy results,rather, we focused on a novel recommendation based approachfor the future ease for a particular data-set for researchers