7 CONCLUSIONS Th e problem of combining classifi ers which use diff erent representations of the patterns to be classifi ed was studied. We have developed a common theoretical framework for classifi er combination and showed that many existing schemes can be considered as special cases of compound classifi cation where all the pattern representations are used jointly to make a decision. We have demonstrated that under diff erent assumptions and using diff erent approximations we can derive the commonly used classifi er combination schemes such as the product rule, sum rule, min rule, max rule, median rule, and majority voting. Th e various classifi er combination schemes were compared experimentally. A surprising outcome of the comparative study was that the combination rule developed under the most restrictive assumptions — the sum rule — outperformed other classifi er combinations schemes. To explain this empirical fi nding, we 3 ACHIEVEMENT/CONTRIBUTION REFINING THE IMPLICATIONS 4 LIMITATIONS CURRENT AND FUTURE WORK APPLICATIONS
Discussion/Conclusion — Writing Task 181 investigated the sensitivity of various schemes to estimation errors. Th e sensitivity analysis has shown that the sum rule is most resilient to estimation errors and this may provide a plausible explanation for its superior performance. Phosphorus removal by chemical precipitation in a biological aerated fi lter DISCUSSION Chemical dosing onto the top of the BAF produced excellent phosphorus removal effi ciencies compared to the removal obtained by biological uptake. Th e performance of the plant was unaff ected with respect to BOD, COD, suspended solids and TKN. In contrast with previous fi ndings, using an aluminium based reagent (Rogalla et al., 1990), the nitrifi cation process was signifi cantly aff ected. Th e use of spent pickle liquor dosing onto an activated sludge plant for phosphorus removal was also seen to aff ect nitrifi cation, especially at high doses (Bliss et al.,1994) although this waste product may contain contaminants toxic to nitrifying bacteria. Th e resulting reduction in nitrifi cation which occurred during chemical dosing of weight ratio 1:1.14 (P:Fe) coincided with the greatest BOD loading (1.74 kg/m3 per day) and the highest NH4 loading (0.4 kg/m3 per day). To achieve complete nitrifi cation an average BOD loading of 1.6 kg/m3 per day has been suggested (Stensel et al., 1988), which was exceeded during this time. At higher BOD loadings the nitrifying bacteria may be outcompeted by the organisms responsible for carbon oxidation (Metcalf and Eddy Inc., 1979) and higher ammonia loadings can create extra pressure for the nitrifying bacteria. Th ese conditions may also explain the increased oxygen demand. A more extensive study of the eff ects of iron dosing on the nitrifi cation process may be required, although the increased BOD loading most likely accounts for the reduction in this process. Th e optimum chemical dose for phosphorus removal is dependant on the EC limit imposed, the stability of the process
182 Science Research Writing required and the capital/running costs available. Unfortunately, specifi c weight ratios could not be studied for any signifi cant period of time due to the variable concentration of incoming phosphorus. Th e use of Fuzzy logic systems (Bulgin, 1994), to adjust the chemical dose with respect to the incoming total phosphorus, would have removed this problem. Overall the most stable and eff ective weight ratio was 1:1.50 (P:Fe). Th is is lower than the optimum ferric chloride dose found previously (Stensel et al., 1988) of 1:2.00 (P:Fe), but a comparison of the performances of iron (II) and iron (III) salts would be useful. Although this produced the most stable effl uent quality, its performance was not signifi cantly diff erent from that produced by a dosing ratio of 1:1.25. Providing the iron (II) solution is changed regularly, because it was at the end of each period when the removal effi ciency deteriorated, a chemical dose ratio of 1:1.25 should be suffi cient to meet EC limits of 1 mg/litre. If, however, a limit of