DRN HYBRID MODEL FOR PREDICTING AUTISM USING RAPID MINER TOOL

: Autism is wrong connection between cells in the human brain which cause abnormalities in the brain structure or function. Every human being with Autism Spectrum Disorder (ASD) has unique symptoms and abilities. Symptoms of ASD typically appear during the first three years of human life. Autism had been classified as three different types such as serve autism, moderate autism, and mild autism. Diagnosing ASD is based on ASD historical dataset because there is no blood or other medical test. With this in mind this paper focuses on developing new hybrid DRN model is created by combining three different models like Deep Learning, Random Forest, and Naïve Bayes (DRN) models. DRN hybrid model is implemented in Rapid Miner tool to find the Accuracy, Precision, recall, Classification error and Executed time. The result obtained shows DRN model is better when compared to the existing models like, Ada Boost, Bagging, Vote, Stacking and Bayesian Boosting models. Hence DRN hybrid can be used to predicting autism using the historical dataset.


I. INTRODUCTION
Autism is a lifelong complex developmental disorder in the human brain. Autism occurs in the early stage of human brain development especially, during the first three years of life. Everyday human brains interpret the things happening in the environment to smell, hear, taste and touch but a child with ASD has problem of interpreting. Every child with autism has unique abilities, wide range of symptoms and challenges. Some symptoms of ASD are lack of communication, social-behavioral challenges and Repetitive behaviors. In this symptoms based Autism has been divided into three different level as serve autism (Low level), moderate autism (medium level), and mild autism (high level). ASD diagnosing is very complex and time consuming because early stage symptoms are found common in adult. For this reason of diagnosing is based on the historical dataset in ASD. In this work historical dataset of ASD is collected from VAERS website. In VAERS website has collected information during the adverse events that occur after administration of vaccine. In this website approximately 30,000 reports are filled and each fields of the report has patient details. presented an evolutionary method for automatic selection of features of speech in a classification task for autism and It is based on a genetic algorithm that selects the best combination of acoustic features using support vector machines as classifier [11]. M.S. Mythili and A.R.Mohamed Shanavas comparisons on data mining classification methods used to predict learning skills of autistic children. The classification algorithms like, J48 and Support Vector Machine (SVM) algorithms are compared check the efficiency and outperform the SVM [12]. Vitthal Manekar and Kalyani Waghmare comparisons on various data mining algorithms like, decision tree, Naïve Bayes, KNN and SVM and Hybrid Approach of SVM algorithm and the SVM algorithm with evolutionary approach is more Accurate [13]. Parvathi I and Siddharth Rautaray describes about the proposal of hybrid data mining model to extract classification knowledge for various diseases in clinical decision system and presents a framework of the tool various tools used for analysis [14]. Priyanka Juneja and Anshul Anand has been analysis on the ASD patient symptom based a pre-level decision is taken about to recognize the probability of ASD [15]. Mohana E and Poonkuzhali.S describes the result for best suited classifier using to predicting the risk level of ASD [16]. Motaz M. H. Khorshid et al. proposed an hybrid machine learning models using to predict terrorist groups responsible of attacks in Middle East and North Africa [18]. Ionuț Taranu describes data mining techniques is most beneficial in health care organization [19]. Priyanka Sanjay Podutwar and Prof. Ms. R. R. Tuteja proposed a soft computing technique using to find the accuracy of ASD [20]. Ashmeet Singh and R Sathyaraj describe various Methodologies used to predicting small and large dataset in rapid miner environment [21].

III. PROPOSED SYSTEM
The implementation work is to predict different level of autism (low, middle, high). Diagnosing initial stage of autism is difficult because early stages of symptoms are very common. Autism diagnosed based on multiple symptoms in the historical dataset. Autism symptoms dataset is collected from VAERS and it is used hybrid model to predict ASD [22]. In Figure 1 represent the process flow architecture of this proposed system. Rapid Miner combines tools and applicability to provide user-friendly integration environment of the newest data mining techniques [17]. Rapid Miner is open-source data mining with java software and platform for data science software. It is provides an integrated environment for data preparation, deep learning, machine learning, predictive analysis and text mining. Rapid Miner is developed on open core model. An application of Rapid Miner covers a wide range of real-world data mining task. Rapid Miner tool supports the DRN hybrid algorithm [23].

B. Dataset:
The raw autism symptoms dataset had been downloaded from the government website of "https://vaers.hhs.gov/data/data" [22]. The ASD dataset is a CSV file format and it is contained eight attribute. Each attribute define autism patient details show in table I.

C. Data Preprocessing:
Data preprocessing is an important steps of this work. The raw dataset cannot be used directly because raw dataset associated with huge amount of unwanted data. Unwanted data is includes Null Values, Redundant Values, and Missing Values. Autism Dataset is removed the unwanted values by the data preprocessing. Preprocessed data will be clear, clean and noiseless.

D. Split Validation:
Split Validation is a nested operation. It has split into two subprocess as training subprocess and testing subprocess. The training subprocess is used for learning or building a hybrid model and trained model is then applied in the testing subprocess. The performance of the hybrid model is also measured during the testing subprocess. Split validation is mainly used to estimate how accurate a hybrid model will perform [23].

E. DRN Hybrid Model:
In this work the proposed hybrid model combined of three models like Deep Learning, Random Forest and Naïve Bayes. The result obtains from the combining DRN Hybrid model outperform in classification parameters than Accuracy, Recall, Precision, Classification-Error, Kappa Statistic and Executed time when implement individual.

a. Deep Learning:
Deep Learning is a set of Machine Learning algorithms which have one or more hidden layers of multiple nonlinear processing units between input and output layers. This is used in the learning phase during training process. The deep learning algorithm makes multiple hidden layers in the autism symptoms dataset [23].

b. Random Forest:
Random Forest in Rapid Miner environment generates a set of random trees. In Random Forest to create a classification model that predicts the value of a label attribute. The resulting of Random Forest model contains a specified number of tress in the ASD dataset [23].

c. Naïve Bayes:
A Naïve Bayes is a probabilistic classifier and it is a prediction models based on Baye's theorem, with focus on independent attributes. Naïve Bayes classifier combines this model with a decision rule. In these model describe the conditional probability of each set of possible causes for a given observed outcome can be computed [23].

IV. RESULT AND ANALYSIS
The New DRN hybrid model used to predict three different level of autism like low, middle and high. In DRN hybrid model is implemented in the Rapid Miner environment to find the Accuracy, Precision, Recall, Classification Error, Kappa Statistic and Executed Time .   In Table II    VI.