Categorical Variable Mapping Considerations in Classification Problems: Protein Application

The mapping of categorical variables into numerical values is common in machine learning classification problems. This type frequently performed a relatively arbitrary manner. We present series four assumptions (tested numerically) regarding these mappings the context protein using amino acid information. assumption involves problems without need to use approaches such as natural language process (NLP). first three relate equivalent mappings, and fourth comparable proposed eigenvalue-based matrix representation chain. These were tested across range 23 different algorithms. It shown that simulations are consistent with presented assumptions, translation permutations, eigenvalue approach generates classifications statistically not from base case or have higher mean while at same time providing some advantages having fixed predetermined dimensions regardless size analyzed protein. generated an accuracy 83.25%. An optimization algorithm also selects appropriate number neurons artificial neural network applied above-mentioned problem, achieving 85.02%. model includes quadratic penalty function decrease chances overfitting.