Results
-
ML model: Collaborative Filtering for Task1
-
When using Collaborative Filtering with Manhattan similarity and k_neighbor = 3 on the random 1/100 dataset, the average RMSE for predicting Purchase in 10-Folds-Cross validation is 3644.533. When implementing Collaborative Filtering with cosine similarity and k_neighbor = 3 on the random 1/100 dataset, the average RMSE for predicting Purchase in 10-Folds-Cross validation is 3361.723.
-
When using Random Forest model with n_estimators = 100 on the random 1/100 dataset, the RMSE for predicting Purchase in 10-Folds-Cross validation is 3123.527. When using Random Forest model with n_estimators = 100 on the entire dataset, the RMSE for predicting Purchase in 10-Folds-Cross validation is 2895.407.
-
ML model: SVM DT KNN for Task2
With SVM, Decision Tree, and KNN and 10 folds cross validation is implemented to assess the algorithm performance. {'gender', 'age', 'occupation', 'city_category', 'stay_in_current_city_years', 'marital_status'} are used as attributes to predict ‘purchase_level’ (from level 1 to level 4), and all attributes are transferred to integer type. Following is the results of 10 folds cross validation.
​
1 2 3 4 5 6 7 8 9 10 Avg
SVM 0.477 0.433 0.513 0.489 0.49 0.503 0.466 0.490 0.491 0.474 0.480
DT 0.385 0.381 0.350 0.330 0.368 0.352 0.387 0.399 0.340 0.346 0.360
KNN 0.338 0.361 0.331 0.333 0.309 0.340 0.362 0.379 0.333 0.320 0.340
​
​