Visualization of Net Effects for Image Hiding Using Gain/Lift

Abstract

At present most of the larger companies depends heavily on their data science capabilities for taking decisions. On the basis of complexity and diversity of analysis, the big data units are transformed into larger and more technologies. Internet technologies are now playing a vital role in our day to day life. It has the advantages along with the disadvantages also, which in term generates the requirements of image hiding technology for maintaining the secrecy of the secret information. The interpretability of findings plays a major role for the success of delivering data science solutions into business reality. Even if the existing method provides outstanding accuracy, they may be neglected if they do not hide the image or text in an exact manner for various cases. When evaluating ML/DL [1] models there is an excess of possible metrics to assess performance. There are things like accuracy, precision- recall, ROC and so on. All of them can be useful, but they can also be misleading or don't answer the question at hand very well. The ROC AUC score is not informative enough for taking decisions since it is abstract for non-technical managers. Hence two more informative and meaningful metrics that every analyst should take into consideration when illustrating the results of their binary classification models: Cumulative Gains and Lift charts. Both the metrics are extremely useful to validate the predictive model (binary outcome) quality. Gain and Lift charts [2] are used to update the performance of binary classification model. They measure how much better one can expect to do with the predictive model. It also helps to find the best predictive model among multiple challenger models. The main intention behind this paper is to assess the performance of the binary classification model and compares the results with the random pick. It shows the percentage of gains reached when considering a certain percentage of the data set with the highest probability to be target according to the classifier. This paper proposes a broad look at the ideas of cumulative gains chart and lift chart to develop a binary classifier model quality which can be used theoretically to evaluate the quality of a wide range of classifiers in a standardized fashion. This paper proposes a hybrid solution of image hiding binary classifier using vicinity value based image hiding classification model as main complimented by gain calculation to increase image hiding classification accuracy. The study has shown that implementing the image hiding binary classification using Gain and Lift is feasible. Experiment of the study has confirmed that the image hiding binary classification model can be improved.