Background: Nowadays, the number of people diagnosed with Alzheimer__ampersandsignrsquo;s disease is increasing rapidly due to lifestyle changes. The average age of Alzheimer__ampersandsignrsquo;s patients is now in the range of 65-80 which was earlier of 75 and above. It has always been challenging to the doctors for proper diagnosis and planning of appropriate treatment of this disease. The manual identification of Alzheimer__ampersandsignrsquo;s disease from MRI images is influenced by different factors and may vary from experts depending on their expertise in the diagnosis of the disease. There is no efficient technique so far for the accurate diagnosis of Alzheimer__ampersandsignrsquo;s disease at the very early stage to prevent the progress of the disease. Objectives: The main aim of the study is to develop an algorithm which could detect the presence of Alzheimer__ampersandsignrsquo;s disease from the brain MRI images by extracting the brain features so that the presence of Alzheimer__ampersandsignrsquo;s disease can be detected at the initial stage itself and appropriate treatment could be given at the right time to prevent its progression. Methods: About 6000 MRI brain images which are used for the study is obtained from ADNI Database. The MATLAB toolbox is used for design and programming. A new algorithm is developed using combined feature extraction using SURF, FAST, BRISK, Harris and Min Eigen methods followed by HOG and feature selection using Principal Component Analysis method for early prediction of various stages of Alzheimer__ampersandsignrsquo;s disease. An analysis of the proposed method is done by combining it with k Nearest Neighbor, Decision Tree and Na__ampersandsigniuml;ve Bayes and Random Forests classifiers and the performance parameters are evaluated. Results: The accuracy of 98.3% is obtained when K-nearest Neighbor is used for classification of Alzheimer__ampersandsignrsquo;s disease. This is followed by Decision Tree having an accuracy of 98.13%, Naive Bayes with an accuracy of 97.31% and Random Forests having an accuracy of 97.12%. Conclusion: The proposed algorithm is found to be superior to the methods which use only one feature extraction method which is developed for the prediction and classification of Alzheimer__ampersandsignrsquo;s disease.