Framework for Improving the Accuracy of the Machine Learning Model in Predicting Future Values

Machine Learning models are generated by providing training data to the learning algorithms. The model designed is rated by the accuracy in predicting future values. Unfortunately, the accuracy of the models is not satisfactory. The presence of different domains requires various data optimization techniques to improve the accuracy of machine learning models. For any predictive analyst, there exists a high correlation between the learning model and the input data. Moreover, the optimization level of the machine learning algorithm is characterized by its speed, accuracy, training time, the volume of data, and implementation level. Hence, for a data scientist, it is a difficult task to decide the granularity of the algorithm on training data considering the issues related to data quality viz, missing values or errors in input data set, presence of textual or Boolean values in data set, an improper number of domains than required, non-uniformity in the features. In general, machine learning systems are considered black boxes even to the research community that builds them. This makes it hard for them to access the results. In this paper, the main focus is laid on testing the quality of training data through a proposed framework. The framework will be designed in line with the aspects of a machine learning model to be tested based on the quality of data, quality of features, and quality of algorithms.