A Comparative Analysis Framework to Classify Drug Data Using Machine Learning and Deep Learning Approach

Abstract

In modern days, Artificial Intelligence (AI) plays a dominant role in almost all sectors like the health sector in classifying patients' data and predicting proper diagnosis and further treatment accordingly. Besides, AI is widely used in weather prediction, financial risk prediction, and stock market prediction and the educational sector to classify students’ performance, to name a few. Machine Learning (ML), a branch of AI, and Deep Learning (DL), a subset of ML, are in great demand to analyze big data obtained from substantial data streams. Different ML/DL techniques are applied on different data sets, irrespective of their size. Theresults obtained depend upon the particular ML/DL technique and their work as per the algorithm built.Predicting the best ML/DL technique for classification or regression or performing clustering depends upon the comparison of results obtained by different classifiers used when performing classification or clustering on a particular data set. In our research work, ML and DL techniques are used in classifying drug reviews based on the topics of drugs present in the data. Existing ML and DL techniques like K-Nearest Neighbor (KNN), Naive Bayes, Random Forest (RF), Logistic Regression (LR), Linear Support Vector Classification (L-SVC), WWR and Deep Neural Network are compared with the proposed Ant Colony Based Deep Belief Neural Network (AC-DBN).A Framework is drawn to compare drug data using the above mentioned ML and DL techniques. AC-DBN and Improved Multinomial Naive Bayes (I-MNB) outperformed by yielding accuracies of 98% and 91%. WWR, KNN yielded 95% and 92% accuracies in classifying drug data, followed by DNN and H-MLA with the precision of 89%. The less complex Classifier-Simple Naïve Bayes gave 87% accuracy, followed by RF, LR and L-SVC with 76%, 74% and 78% accuracy. Furthermore, different ML and DL techniques are compared on other evaluation parameters in terms of precision, recall and f-score